APPROVED: Lawrence F. Williams, Major Professor Marijn S. Kaplan, Committee Member Marie-Christine W. Koop, Committee
Member and Chair of the Department of Foreign Languages and Literatures
Sandra L. Terrell, Dean of the Robert B. Toulouse School of Graduate Studies
FROM "Y AS PLUS PERSONNE QUI PARLE" TO "PLUS PERSONNE NE DIT RIEN":
THE VARIABLE USE OF THE NEGATIVE PARTICLE NE
IN SYNCHRONOUS FRENCH CHAT
Rémi A. van Compernolle, B.A.
Thesis Prepared for the Degree of
MASTER OF ARTS
UNIVERSITY OF NORTH TEXAS
May 2007
van Compernolle, Rémi A., From "y as plus personne qui parle" to "plus
personne ne dit rien": The variable use of the negative particle ne in synchronous
French chat. Master of Arts (French), May 2007, 114 pp., 30 tables, 3 figures,
references, 62 titles.
This study analyzes negative particle variation (i.e., the variable presence or
absence of the negative particle ne) in synchronous French chat discourse within a
labovian-inspired framework. Selected morphosyntactic, lexical, and phonological
constraints are considered. Multivariate analyses performed by GoldVarb 2001
revealed that subject type (i.e., NP, [- overt] subject environment, pronoun) and the
phonological environment preceding the position of ne—regardless of its presence or
absence—are determining factors in the variation. In addition, discursive-pragmatic
effect was explored in a sub-sample of data. The results indicate that ne is seldom
present in verbal negation during explanatory discourse style, yet it is very likely to be
retained in ludic, emphatic, and proverbial styles.
ii
Copyright 2007
by
Rémi A. van Compernolle
iii
TABLE OF CONTENTS
Page
LIST OF TABLES.........................................................................................................vi LIST OF FIGURES .................................................................................................... viii Chapter
1. INTRODUCTION ..................................................................................1
1.1 Aim and Scope............................................................................1
1.2 Formal Explanation and History of Verbal Negation in French.3
1.2.1 Old French and Middle French .......................................3
1.2.2 Classical and Modern French..........................................5
1.2.3 Pleonastic ne ...................................................................6
1.2.4 Summary .........................................................................7
1.3 Theoretical Framework...............................................................8
1.3.1 The Variationist Tradition...............................................8
1.3.2 Stylistic Variation and Register ....................................11
1.3.3 Speech Communities ....................................................12
1.3.4 Applying a Variationist-Inspired Framework to Synchronous CMC........................................................13
1.4 Research Questions...................................................................14
1.5 Organization of the Thesis ........................................................15 2. REVIEW OF THE RELEVANT LITERATURE ................................16
2.1 Overall Retention Rates in Previous Studies ............................16
2.2 Social Factors............................................................................18
2.3 Phonological Factors.................................................................21
2.4 Linguistic Factors......................................................................22
2.5 Stylistic Factors.........................................................................29
2.6 Summary of the Relevant Literature.........................................31 3. METHODOLOGY ...............................................................................33
3.1 The IRC Environment...............................................................33
iv
3.2 Data Collection .........................................................................35
3.3 The Concordance© Software and Counting Methodology ......36
3.4 Examples of Negation Found in the Corpus .............................38
3.5 Coding of Tokens......................................................................41 4. ANALYSIS OF THE CORPUS ...........................................................43
4.1 Distribution of 2Neg and 1Neg.................................................43
4.2 GoldVarb 2001 and Statistical Procedures ...............................45
4.3 Linguistic Factors......................................................................47
4.3.1 Analysis of First Coding ...............................................47
4.3.2 Analysis of Second Coding: Phonological Environment .......................................................................................49
4.3.3 Analysis of the Second Coding: Morphosyntactic and Lexical Environment.....................................................52
4.3.4 Interaction of Subjects and Phonological Environment .......................................................................................59
4.3.5 Summary of Results: Second Coding ...........................61
4.4 Discursive-Pragmatic Effect .....................................................62
4.4.1 Overview of Style-Shifting ...........................................62
4.4.2 Overview of Selected Participants ................................65
4.4.3 Definitions and Examples of Discourse Styles.............66
4.4.3 Results...........................................................................69 5. CONCLUSION.....................................................................................72
5.1 Summary of Results..................................................................72
5.2 Direction for Future Research...................................................75 Appendices
A. GOLDVARB 200 CODING KEY (ALL DATA) ................................76 B. ONE-STEP BINOMIAL: FIRST CODING .........................................78 C. STEP-UP/STEP-DOWN ANALYSIS: FIRST CODING ....................81 D. ONE-STEP BINOMIAL: SECOND CODING (PHONOLOGICAL
ENVIRONMENT)................................................................................87 E. STEP-UP/STEP-DOWN ANALYSIS: SECOND CODING
(PHONOLOGICAL ENVIRONMENT) ..............................................89
v
F. ONE-STEP BINOMIAL: SECOND CODING (MORPHOSYNTACTIC ENVIRONMENT) ......................................92
G. STEP-UP/STEP-DOWN ANALYSIS: SECOND CODING
(MORPHOSYNTACTIC ENVIRONMENT) ......................................94 H. GOLDVARB 2001 CODING KEY (SELECTED DATA)..................98 I. GOLDVARB 2001 RESULTS FOR SELECTED DATA:
PHONOLOGICAL ENVIRONMENT...............................................100 J. GOLDVARB 2001 RESULTS FOR SELECTED DATA:
MORPHOSYNTACTIC ENVIRONMENT AND DISCOURSE STYLE ................................................................................................104
REFERENCES ...........................................................................................................109
vi
LIST OF TABLES
Page
1. Development of Verbal Negation in French.......................................................7
2. Recent Decline of French ne in France.............................................................17
3. Age of Speaker and ne Retention......................................................................19
4. Retention Rates of ne Reported in Previous Studies According to Neg2 Type ...........................................................................................................................23
5. Ne Retention Rates in Preformed Sequences....................................................25
6. Ne Retention According to Subject Type as Reported in Previous Studies .....26
7. Ne Retention in [- overt] Subject Environments as Reported by Coveney (1996) and Armstrong & Smith (2002) ............................................................28
8. Turns and Human-Generated Words ................................................................36
9. Types of 1Neg Found in the Corpus .................................................................38
10. Distribution of 2Neg and 1Neg.........................................................................43
11. 1Neg Distribution..............................................................................................44
12. Overall Distribution of 1Neg Type D and 2Neg...............................................44
13. One-Level Binomial: First Coding ...................................................................48
14. One-Level Analysis of Phonological Environment Factor Weights: Second Coding...............................................................................................................50
15. 2Neg and 1Neg Distribution According to Preceding Phonological Environment......................................................................................................50
16. 2Neg and 1Neg Distribution According to Following Phonological Environment......................................................................................................51
17. Ne Retention According to Surrounding Phonological Environment...............51
18. Interaction of Selected Phonological Environments.........................................52
19. Factor Weights of One-Level Binomial Analysis for Neg2 Type, Subject Type, and Sentence Type: Second Coding .......................................................53
vii
20. Distribution of 1Neg Type D and 2Neg According to Neg2 Type: Second Coding...............................................................................................................54
21. Single vs. Multiple Neg2s: Second Coding ......................................................54
22. Ne Retention According to Sentence Type: Second Coding ............................55
23. Distribution of 2Neg and 1Neg Type D According to [+ overt] Subject Type: Second Coding ..................................................................................................56
24. Distribution of 2Neg and 1Neg Type D with Clitic and Non-clitic Subject Pronouns: Second Coding.................................................................................57
25. Distribution of 2Neg and 1Neg Type D According to Subject Type: Second Coding...............................................................................................................58
26. Pronoun with Final Vowel and Following Phonological Environment............60
27. Pronoun with Final Consonant and Following Phonological Environment .....60
28. Distribution of 2Neg and 1Neg Type D for Top Six Contributors ...................66
29. Distribution of 2Neg and 1Neg Type D According to Subject Type ...............70
30. 2Neg and 1Neg Type D Distribution According to Discourse Style................70
viii
LIST OF FIGURES
Page
1. Concordance© Software: Headword List and Occurrences .............................36
2. Concordance© Software: Text Viewer.............................................................37
3. Bell's Model of Audience Design .....................................................................64
CHAPTER 1
INTRODUCTION
1.1 Aim and Scope
The emergence of various new technologies has enabled communication to occur in a
variety of new social contexts "through the medium of written language" (Werry, 1996, p.
47). However, communication in electronic environments is often different from
communication in more traditional contexts since interlocutors do not see or hear one another
and therefore "do not have access to non-verbal information about how others are
responding" (Herring, 1999). Nonetheless, much of the literature on Computer-Mediated
Communication1 (CMC) attempts to compare discursive, communicative, and linguistic
features of electronic discourse with those found in either written or spoken language.
CMC can be divided into two broad categories: asynchronous CMC (e.g., electronic
mail, discussion forums,2 etc.) and synchronous CMC3 (e.g., public chat rooms and instant
messaging services). The language used in asynchronous CMC is often likened to that of
more traditional forms of written language (e.g., written letters, newspapers, literature, etc.),
while the language used in synchronous CMC often appears to resemble everyday
conversational speech,4 at least to some extent.
1 The more recent—and more inclusive—label Computer-Mediated Discourse (CMD) might be more appropriate, but Computer-Mediated Communication is used in this thesis. See Herring (2001) for an overview of CMD. 2 I have used "forums" instead of "fora" since my informal observations of the lexicon used on the Internet suggest that "discussion forums" is the preferred plural form. 3 Garcia & Jacobs (1999) proposed the term "quasi-synchronous" since the recipient of the message must wait for the sender to complete his or her message and press "send" or "enter" before having access to the content of the message. 4 One exception to this comparison is moderated chat (Williams, 2006; van Compernolle & Williams, in press).
1
Although communication researchers and linguists have studied many aspects of
CMC, there exists relatively little empirical data concerning specific linguistic traits and
variables that have been explored extensively in more traditional communicative
environments (e.g., written and spoken discourse). Since CMC environments—especially
synchronous text-based chat (see Collot & Belmore, 1996; Werry, 1996; Anis, 1999; Dejond,
2002; Pierozak, 2003a)—are, to say the least, becoming increasingly popular in parts of the
world where access to networked technologies is widespread, it follows that the observation
and documentation of specific linguistic features of discourse used in this context could prove
rather insightful and have the potential to inform research in linguistics and communication
studies, among other fields. The present study aims to explore one of the most well known
grammatical variables in the modern French language: the use of the negative particle ne.
Although the ne paradigm (i.e., ne present vs. ne absent) has been studied in spoken
French by many linguists over the course of the past thirty-five years, it has not yet been
explored in synchronous, French-language CMC. 5 The data in this study suggest that ne use
in synchronous, French-language CMC discourse is very similar to ne use in informal spoken
discourse; that is, ne is often omitted from verbal negation, yet it remains an important
sociolinguistic resource in communication since its presence usually correlates with a number
of stylistic and pragmatic features of discourse.
In addition to comparing the extent to which French language chat participants use ne
in a variety of syntactic, stylistic, and pragmatic environments with results reported in similar
studies on spoken French, this study highlights a number of discursive features particular to
French-language synchronous CMC that co-occur and co-vary with ne. In this way, the effect
of these features of CMC discourse on the ne paradigm will be explored.
5 I have previously reported various findings of this research at the Colloque International: La Langue de la Communication Médiatisée par les Technologies de l'Information et de la Communication in Bordeaux, France in May 2006 and at the 2006 Conference of the Association for French Language Studies in Bristol, England in September 2006 (with Lawrence Williams).
2
1.2 Formal Explanation and History of Verbal Negation in French
In prescriptive Modern French, verbal negation is achieved by combining the proclitic
(i.e., pre-verbal particle) ne and another word or adverb that has a negative meaning. When a
conjugated verb form is used, ne precedes the verb, as well as its pronominal complements,
and the second-negative (Neg2) follows (Grevisse, 1993). This type of negation will be
referred to as two-particle negation (2Neg).
2Neg has not, however, always been required in the French language. Indeed, verbal
negation in French has undergone a number of developments and changes throughout the
history of the language. In order to understand the current state of negation in French, the
following paragraphs provide a summary of the development of 2Neg.6
1.2.1 Old French and Middle French
French—like a number of other European languages—inherited much of its grammar
and syntax from Latin, including the negation non (Brunot, 1966; Dauzat, 1953, 1964; Pope,
1961; Ewert, 1969; Rohlfs, 1970). In Vulgar Latin, which was spoken throughout much of
France into the Middle Ages, non could be used alone with a conjugated verb form to express
negation (Sancier-Chateau, 1993, p. 93). However, as the French language distinguished
itself from Vulgar Latin, non weakened to nen in pre-verbal position and eventually to ne (or
n' immediately preceding a vowel). Accented non was restricted to elliptic use in negative
responses to questions and a certain number of archaisms (Rohlfs, 1970; Sancier-Chateau,
1993).7
In Old French, "ne constituted sufficient negation in itself" (Rickard, 1989, p. 54) and
a variety of words (e.g., pas, point, mie, etc.) could be added for emphasis. Toward the end of
the 15th century, the words pas and point emerged as the default second-negatives (Dauzat,
1967; Ewert, 1969), but their use remained emphatic, and they were often omitted in favor of
6 For a general treatment of negation in French, see Pohl (1968) and Kayne (1983). 7 For a general treatment of phonology and morphology in Old French, see Pope (1961) and Rohlfs (1970).
3
single-particle negation with ne. According to Cohen (1967), two-particle negation had
become widespread by the 16th century; however, this usage had not yet become regular and
single-particle negation was still preferred (Brunot, 1966, v. 2, p. 472).
As second-negatives were used more frequently in verbal negation, they began to lose
their emphatic quality and "gradually . . . came to be invested with a negative meaning"
(Ewert, 1969, p. 260). Ashby (1981) has referred to the grammaticalization of second-
negatives as "an innovation in French" since, in other Romance languages, "non and its
descendents are only sporadically reinforced, especially by nouns denoting smallness or
insignificance" (p. 674). Dauzat (1967) offered the following explanation for such a
development.
Toutes les langues romanes ont éprouvé le besoin de renforcer la négation qui
accompagne le verbe; mais c'est en français que cette tendance s'est
développée au maximum, surtout parce que non, atone, s'était affaibli en nen,
puis ne, n', particule phonétiquement trop faible pour exprimer une négation
énergique. (p. 196)8
By the 16th century, second-negatives had become so common and understood as
negative that ne was often omitted in direct interrogatives involving subject-verb inversion
(e.g., Vient-il pas; see Brunot, 1966, v. 2; Sancier-Chateau, 1993). However, by the 17th
century "les théoriciens cessent de considérer pas comme suffisant dans les interrogations
directes" (Brunot, 1966, v. 4, p. 1,039),9 and two-particle negation was recommended.
In addition, grammarians began to consider two-particle negation as required in all
instances of verbal negation. For example, François de Malherbe (official poet of the court
under Henri IV and later Louis XIII) was of the opinion that "the negative consist[ed] of two 8 Translation: "All Romance languages have experienced the need to reinforce the negation that accompanies the verb; but it is in French that this tendency was developed the most, especially because non, an atonic [unstressed or unaccented] syllable, had weakened to nen, then ne, n', a particle that was too phonetically weak to express an energetic negation." 9 Translation: "theorists cease to consider pas as sufficient in direct interrogatives."
4
parts, not ne alone, but with pas or point" (Rickard, 1989, p. 102), and it is clear that this was
the preferred structure of the time. However, the second-negative was still omitted fairly
often well into the 17th century (Cohen, 1967). Brunot (1966, v. 4) commented, however, that
"la suppression de pas est en général dans le style marotique, ce qui achève de donner à cet
archaïsme son caractère" (p. 1,034).10
Another question arose concerning the placement of the second-negative with
infinitives and, by extension, object pronouns. Until the 17th century, the second-negative had
always followed the verb, even with infinitival forms. However, "[a]u fur et à mesure que
pas, point devenaient les compléments indispensables de ne, ils devaient, suivant une loi
invariable, s'en rapprocher" (Brunot, 1966, v. 4, p. 1,039),11 and pas was placed in front of
the infinitival form. Pronouns, on the other hand, were often placed between the two negative
particles.
1.2.2 Classical and Modern French
According to Rickard (1989), it was not until the 18th century that the rules of
negation became fixed. The Académie française decided that two-particle negation was
required in direct interrogatives, and two-particle negation (especially ne...point) became the
standard. Ne was seldom used alone; this usage was limited to a few select verbs (e.g.,
pouvoir, savoir, empêcher) and in a limited number of syntactic environments (Grevisse,
1993).
Negation in Modern French follows Classical French usage, although pas has replaced
point as the most common second-negative (Dauzat, 1967). The two negative particles
"embrace" (i.e., surround) the conjugated verb or, in the case of an infinitival form, they
precede the verb and its pronouns. Simple negation with ne is, however, still in usage, albeit
10 Translation: "the suppression of pas is in general Marotian in style, which gives this archaism its character." (Marotian refers to the writing style of the 16th century French poet Clément Marot.) 11 Translation: "as pas, point became the indispensable complements of ne, they had to get closer, following an invariable law."
5
in a very limited number of contexts, one of which is proverbs (see Grevisse, 1993, p. 1,448,
§ 974).
According to Dauzat (1954) each part of negation has a specific function: "le premier
élément [ne] énonce une discordance, complétée par le second [pas, point, rien, etc.] qui
exprime la forclusion. La porte s'ouvre par une particule qui prépare la négation, le verbe
passe, et le vantail se referme sur une précision de la valeur négative" (p. 218).12
Nonetheless, it appears that negation is expressed principally by the second particle,
rendering ne somewhat redundant (Ashby, 1981).
The weakening of ne and the strengthening of second-negatives has led to a new
development in modern (spoken, primarily informal or everyday) French: single-particle
negation with the second-negative alone. Ewert (1969) noted that "ne, being a mere proclitic
and incapable of bearing a stress, is . . . weakened and tends to be omitted as unessential" (p.
260). This usage has not, however, been well received by purists of the French language.
Dauzat (1954) argued, for example, that "en concentrant la négation sur un seul mot, on
enlève au français une de ses élégances, une finesse d'expression propre à notre langue et que
nous devons avoir à cœur de conserver" (pp. 218-219).13 Nonetheless, it is clear that "[d]ans
la langue parlée, surtout familière, le ne disparaît avec des fréquences variables" (Grevisse,
1993: p. 1462, § 982b, bold in original).14
1.2.3 Pleonastic ne
Dauzat (1954) has argued that pleonastic ne ("ne explétif"; e.g., ne...Ø) is still an
important part of the modern French language. Although single-particle negation with ne is
not usually considered to be a complete negation in modern French as it was in Old and
12 Translation: "the first element announces discordance, completed by the second which expresses the debarment. The door is opened by a particle that prepares the negation, the verb passes through and the door is closed on a precision of the negative value." 13 Translation: "by concentrating the negation on a single word, one removes from French one of its elegances, a finesse of expression particular to our language and that we must be committed to conserve." 14 Translation: "in the spoken language, especially informal, ne disappears at variable frequencies."
6
Middle French, it can be used to help express a number of nuances.
Dans les propositions dites complétives, ne exprime la crainte . . . , la
précaution . . . , l'empêchement. . . . Loin d'être explétive, comme le croyait à
tort l'ancienne grammaire, cette particule a une valeur affective très nette et
permet, là où un flottement est possible, d'exprimer des nuances, ainsi entre
"avant qu'il vienne", plus positif, et "avant qu'il ne vienne", qui entrebâille la
porte au doute. (Dauzat, 1954, p. 218)15
This usage does not appear to be widespread, and it is usually associated with formal or
literary discourse. Indeed, Grevisse (1993) considers this type of ne use to be optional.
1.2.4 Summary
The development of verbal negation in French can be divided into four distinct stages,
as shown in Table 1.
Table 1. Development of verbal negation in French16 Period Negation Old French ne + verb Middle French ne + verb (+ second-negative) Classical French ne + verb + second-negative Modern French (ne) + verb + second-negative
In Old French, ne is used alone in verbal negation. Later, ne is sometimes reinforced
with another word or adverb in Middle French. In Classical French, both ne and a second-
negative are required, but as Modern French develops, ne is sometimes omitted in speech.
Although it is not possible to hypothesize at this point whether ne will disappear from the
French language in the future (Ashby, 1981; Hansen & Malderez, 2004), there is evidence to
support that its use is becoming increasingly dependant on a number of social, linguistic, and
stylistic factors. In other words, the presence or absence of ne is variable, and the variation
15 Translation: "In completive clauses, ne expresses fear, precaution, impedance. Far from being expletive, as the traditional grammar wrongly believed, this particle has a very clear value and allows, where a hesitation is possible, the expression of nuances, such as between 'avant qu'il vienne', more positive, and 'avant qu'il ne vienne', which holds the door open to doubt." 16 Adapted from Ashby (1981).
7
can be linked to various aspects of the speaker's social identity, the formality of the
communicative context, the syntactic environment surrounding the ne position, and a number
of pragmatic features of discourse, among other factors.
1.3 Theoretical Framework
Before continuing to a review of the relevant literature concerning ne use in modern
French, an explanation of the underlying theoretical notions and principles used for the
analysis of the corpus is sketched in the following sections.
1.3.1 The Variationist Tradition
No language on earth is invariable, and there is no monolithic speaker of any language
(Sax, 2003). Indeed, variations in language occur constantly. Labov (1972) noted that:
Most such variations occur only once, and are extinguished as quickly as they
arise. However, a few recur, and, in a second stage, they may be imitated more
or less widely, and may spread to the point where the new forms are in
contrast with the older forms along a wide front. (p. 2)
In the present study, I am operating under the hypothesis that such variations (i.e., those that
recur and spread) in any given language are the direct result of social interactions.
The notion that language variation results from social interaction was first advanced in
the 1960s and 1970s by Labov, according to whom, "one cannot understand the development
of a language change apart from the social life of the community in which it occurs. Or to put
it another way, social pressures are continually operating upon language, not from some
remote point in the past, but as an immanent social force acting in the living present" (1972,
p. 3). Labov (1972) provided three properties of a linguistic variable:
First, we want an item that is frequent, which occurs so often in the course of
undirected natural conversation that its behavior can be charted from
unstructured contexts and brief interviews. Secondly, it should be structural:
8
the more the item is integrated into a larger system of functioning units, the
greater will be the intrinsic linguistic interest of our study. Third, the
distribution of the feature should be highly stratified: that is, our preliminary
explorations should suggest an asymmetric distribution over a wide range of
age levels or other ordered strata of society. (p. 8)
This type of variation "presuppose[s] the option of saying 'the same thing' in several different
ways: that is, the variants are identical in referential or truth value, but opposed in their social
and/or stylistic significance" (Labov, 1972, p. 271).
The speaker's age, gender, and social class are, among many others, some of the social
factors that appear to influence linguistic variables. In addition, a certain number of internal
and external linguistic factors—such as clause and sentence type, phonological environment
and attention paid to speech—have been shown to be influential. These notions have been
defended, critiqued, and reformulated by numerous sociolinguists (Lavandera, 1978; Sankoff,
1980; Bell, 1984; Wolfram, 1991; Coveney, 1996; Eckert & Rickford, 2001; among others)
over the past 30 years.
Although Labov's work was principally concerned with phonological variation, his
theory and method have since been applied to the study of syntactic, morphological, and
lexical variation. The application of variationist theory to non-phonological variation has not,
however, gone without criticism. Lavandera (1978) warned that "it is inadequate at the
current state of sociolinguistic research to extend to other levels of analysis of variation the
notion of sociolinguistic variable originally developed on the basis of phonological data" (p.
171). However, she does not dismiss the idea of applying such analysis to non-phonological
data; rather, she assigns "a different status to such data because they need further
interpretation" (p. 3).
9
Sankoff (1972; discussed in Lavandera, 1978) was one of the first to suggest that the
variationist framework could be applied to non-phonological data. She posited that "the
extension of probabilistic considerations from phonology to syntax is not a conceptually
difficult jump" (p. 58). Her analysis was supported by three examples of non-phonological
variation: the placement of the future marker in New Guinea Tak Pisin bai and, in Montreal
French, the use of the complementizer que and indefinite on.17 Sankoff's study supported the
call for the application of the variationist framework to "cases in which the variation seem[s]
not to be the carrier of social and stylistic significance" (Lavandera, 1978, p. 173). In her
conclusion, however, Lavandera (1978) reiterates her reticence to draw a parallel between
syntactic alternation and sociolinguistic variation, unless the following conditions hold:
(1) that [the variables] can be proven to be carriers of non-referential
information, to have social and stylistic or other significance . . . and (2) that
they prove to be a kind of device of the language similar to phonological
variables, that is, elements whose defining property is a quantifiable
covariation and for which the frequency relationships are the very signals of
those differences. (p. 181)
Milroy (1987) added that the study of syntactic variation can be problematic since
"the difficulty [is not] in obtaining tokens of a variable, but in obtaining the full range of
realizations associated with it" (p. 144, italics in original). It is, however, possible to obtain
the "full range of realizations" of the variable ne; that is, in a negative sentence, ne is either
absent or present. Ne is clearly an example of a syntactic variable that can be analyzed within
a framework inspired by the variationist tradition.
17 Que, used as a complementizer (i.e., the subordinating conjunction que), is sometimes omitted in informal speech (e.g., je pense [que] c'est une bonne idée 'I think [that] it's a good idea'). On is the default indefinite pronoun in French, yet it is being replaced by the second-person pronouns tu/vous. On is also used as the first-person plural pronoun at the expense of nous. For a general treatment of the pronouns on, tu, and vous, see Peeters (2006).
10
1.3.2 Stylistic Variation and Register
Although this study does not specifically address stylistic variation and register (i.e.,
language use in two or more different communication environments is not being compared), I
have provided a brief overview of these notions.
In general, every speaker of every language is thought to have at his or her disposal a
variety of speech styles, each of which is judged appropriate or not for a given
communicative context (Labov, 1970 & 1972; Trudgill, 1974; Hymes, 1984; Sax, 2003). This
is not only true for phonological variables, but for syntactic, morphological, and lexical
variations as well.
Halliday (1976) argued that grammar is composed of a system of choices; in other
words, "[t]he speaker of a language, like a person engaging in any kind of culturally
determined behaviour, can be regarded as carrying out, simultaneously and successively, a
number of distinct choices" (p. 3). The key notion to remember is that language use is
"culturally determined." In addition, language choices (i.e., which forms are used when) are
not free; rather they are dependent upon a number of factors, including communicative
environment, perception of formality, and attention paid to speech (Labov, 1972).
Presumably, as speakers move from one context to another, they shift or modify their style of
speech so that it may be appropriate for their interlocutors (Bell, 1984, 2001).
Speech style, as defined by Labov (1972) and Bell (1984, 2001), is often used
synonymously with the term register (Sax, 2003). For the purposes of the present study, I
wish to distinguish register from style. I have adopted Biber & Finegan's (1994) definition of
register. Broadly defined, a register "is a language variety viewed with respect to its context
of use" (Biber & Finegan, 1994, p. 4). According to Fischer (1958) levels of formality—both
perceived and real—are often associated with registers of language. Register, then, can be
regarded as a language variety that is dependent upon the level of formality perceived to be
11
appropriate in a given communicative context. I will therefore save the term style to refer to
different ways of speaking within the same register. It will be shown later that several styles
can exist at the same level of formality and the variable use of ne depends in large part on
which style is used.
1.3.3 Speech Communities
Much of the sociolinguistics literature (Labov, 1972; Milroy, 1987; Biber & Finegan,
1994; Chambers, 2003) makes use of the term speech community to describe a group of
people that share a common set of linguistic behaviors. According to Labov (1972):
The speech community is not defined by any marked agreement in the use of
language elements, so much as by participation in a set of shared norms; these
norms may be observed in overt types of evaluative behavior, and by
uniformity of abstract patterns of variation which are invariant in respect to
particular levels of usage. (pp. 120-121)
Recent research (Pierozak, 2003a, 2003b & 2003c, van Compernolle, 2006; Williams,
2006; Williams & van Compernolle, 2007; van Compernolle & Williams, in press) has
suggested that CMC users participate in a number of shared norms, some of which are closely
related to norms found in non-electronic environments, while other norms have developed
specifically in and for electronic environments. It appears that "the lack of geographical
constraints and the ability to self-select one's on-line community and communication
environment have played very important roles in defining many of the norms for behaviors
and practices of Internet users" (Williams & van Compernolle, 2007, p. 815).
The definition of community has changed since the advent of the Internet and other
networked technologies. According to Mosco (2004), "existing communities are strengthened
and whole new 'virtual' communities arise from the creation of networks of people who share
interests, commitments, and values" (p. 31). In addition, new virtual communities in which
12
people chose to participate might be better than traditional ones that are constructed by
geographical proximity rather than shared interests, commitments, and values since "a
community that hasn't been chosen is a community of lesser quality" (Whittle, 1997, p. 240).
Some have even argued that networked technologies allow people to form "smaller, more
caring communities" (Hearn et al., 1998, pp. 62-63).
The freedom to choose one's acquaintances and terms of participation is one of the
most important aspects of on-line communities (Whittle, 1997). "[E]ach participant has the
choice to stay or to leave; therefore, those who stay have chosen to abide by whatever norms
have been established by the community as a whole" (Williams & van Compernolle, 2007, p.
38). Since synchronous CMC occurs through the medium of written language, these
communities are essentially linguistic (Pierozak, 2003a); therefore, the norms that have been
constructed by participants in these on-line communities are, in general, linguistic in nature
(Whittle, 1997). It follows that one might consider on-line communities to be speech
communities that exist in electronic environments, and whose social norms consist of
principally written linguistic behaviors.
1.3.4 Applying a Variationist-Inspired Framework to Synchronous CMC
One of the major limitations of variationist studies in the past has been the "observer
paradox" (Labov, 1972). Since the informant is conscious of the presence of the observer
(regardless of how informal or familiar the context may be), it is uncertain that the
interviewee's style reflects that which he or she uses when the observer is absent. In the
present study, this limitation appears to be minimized.
Synchronous CMC offers a rather anonymous communication environment that
allows data collection to take place without the informants being aware of the observer.
Moreover, informants are speaking to one another, which is similar to the group session
method of observation. This type of observation usually reveals more about the vernacular of
13
the speakers than one-to-one interviews since informants are not directly addressing the
observer (Labov, 1972). It follows that the language observed in the present study is
representative of the vernacular of the synchronous CMC environment.
One limitation must, however, be noted. The anonymity provided by this form of
communication prevents observers from gathering (credible) demographic data about their
informants. It is therefore impossible to explore variation according to informant's gender,
age, or social class. However, research has shown that, at least in France, the average
synchronous CMC participant is young (under 25) and belongs to the middle or upper-middle
class (Pierozak, 2003c).
In light of this limitation, the analysis will focus on a variety of linguistic and stylistic
factors that may co-occur and co-vary with the use of ne in synchronous CMC. In addition, a
certain number of pragmatic features particular to synchronous CMC discourse will be
examined.
1.4 Research Questions
The present study is motivated by two objectives. First, this research aims to
contribute to existing scholarship concerning the variable ne in modern French. To this end,
the distribution of two-particle negation (2Neg) and single-particle negation (1Neg) will be
explored in a variety of grammatical, phonological, and pragmatic environments. Second, this
study aims to distinguish chat discourse as one type of modern French that requires more
attention from linguists; therefore, I will discuss briefly a number of discursive features
particular to synchronous electronic environments.
The present study will focus on four general questions:
(1) How do overall rates of ne retention in chat discourse compare to those reported in
studies of spoken French?
14
(2) How do selected linguistic factors (e.g., Neg2 type, subject type, phonological
environment, etc.) favor or disfavor ne retention?
(3) To what extent is the use of ne influenced by sociopragmatic features of discourse
(e.g., emphasis, humor, etc.)?
(4) Which features of discourse particular to chat—which may differ from those
observed in spoken, primarily informal, discourse—seem to influence ne use in this
type of communication?
1.5 Organization of the Thesis
In Chapter 1, I have presented the aim and scope of the present study and defined the
notions and analytical framework to be used. I have also provided a formal explanation of the
history of ne and Neg2 usage in the French language. In Chapter 2, I review relevant
literature, and in chapter 3, I provide a description of the data collection method, as well as
the counting and coding procedures that I have used. In chapter 4, I present my results in
comparison with results reported in previous studies of negation in spoken French. Finally,
chapter 5 includes a discussion of my results, and the research questions are directly
addressed. In addition, I have included a discussion of a certain number of sociopragmatic
features of chat discourse that influence the use of ne and other linguistic variables that
require further investigation.
15
CHAPTER 2
REVIEW OF THE RELEVANT LITERATURE
The ne paradigm is, according to Coveney (1996), "possibly the best known
sociolinguistic variable in contemporary French" (p. 55). Much has been written on negation
in European and Canadian French, and it seems that ne use is under the influence of a certain
number of social factors, including the speaker's age, gender, level of education, and social
class. In addition to these social factors, ne use is very much dependent on a variety of
phonological, syntactic, and pragmatic features of discourse.
In the following sections, I review results reported in a number of studies of ne in
modern French. The review is organized by influence (e.g., social factors, phonology,
syntactic constraints, etc.) instead of chronology of when studies were done. This type of
review provides a clearer picture of the complex of factors that affect the ne paradigm.
2.1 Overall Retention Rates in Previous Studies
A number of studies (Ashby, 1981, 2001; Coveney, 1996; Armstrong & Smith, 2002;
Hansen & Malderez, 2004) have suggested that ne use is in decline throughout much of
France. Although it is unclear whether ne will disappear altogether from the French language
in the future, it is clear that there has been a continuous tendency to omit ne in spoken French
at higher frequencies over the course of the past half century. Table 2 illustrates this tendency
according to the results reported in a number of recent studies on the variable use of ne.
16
Table 2. Recent decline of French ne in France.
Corpus Year of Data
Collection Location Overall
Retention Rate Ashby (1981) 1976 Tours, France 37.0% Coveney (1996) 1989 Somme, France 18.8% Hansen & Malderez (2004) 1989-1993 Ile-de-France/Oise, France 8.2% Ashby (2001) 1995 Tours, France 18.0% Armstrong & Smith (2002) 1997 France 72.5% Although Armstrong & Smith's (2002) study seems to contradict the hypothesis that
ne use is declining in modern French, it is important to note that they explored ne retention
rates in radio shows recorded in 1997 (interviews with politicians, celebrities, etc.), with a
similar corpus from 1960-61 and did find a noticeable decline in ne retention rates. The
relatively high rate of ne retention is most likely attributed to the serious nature of the radio
programs; as a result, the interviewees and hosts may have felt the need to show a higher
level of formality, which often appears to favor ne retention. Nonetheless, Armstrong &
Smith (2002) concluded that the results of their study "suggest strongly that ne deletion is
spreading to highly monitored speech styles" (p. 39). The other studies constitute recorded
conversations between the informants and researchers that took place in more informal,
conversational contexts (i.e., less highly monitored), which often seems to disfavor ne
retention.
The difference between ne retention rates in Ashby (1981) and (2001) clearly shows
that French speakers in the city of Tours have begun to omit ne at higher frequencies. Hansen
& Malderez (2004) demonstrate this trend as well, comparing a corpus of spoken French
recorded by Péretz-Juillard between 1972 and 1974 (see Péretez-Juillard, 1977) in the Paris
area. Rates dropped from 15.8% in the Péretz-Juillard corpus to only 8.2% in the Hansen &
Malderez corpus. Finally, as mentioned above, Armstrong & Smith (2002) also found that ne
retention rates had dropped even in formal, highly monitored speech contexts where one
might expect ne to be retained more frequently.
17
Although results reported in the three diachronic studies (Ashby, 2001; Armstrong &
Smith, 2002; Hansen & Malderez, 2004) demonstrate that ne deletion is becoming more
acceptable at many levels of society and in a variety of social contexts, ne does not seem to
be disappearing from the French language; rather, its use is "dependent on internal linguistic,
stylistic, and social factors" (Ashby, 1981, p. 686; see also Hansen & Malderez, 2004).
Although Coveney (1996) did not conduct a diachronic study of ne retention, his
results demonstrate that ne is used at a rather low frequency in his corpus, which focused on
many social and linguistic factors that had previously been shown to influence ne use (e.g.,
Ashby, 1981). These factors will be discussed in the following sections as they have been
observed and documented in previous studies.
2.2 Social Factors
Previous research (Ashby, 1981, 2001; Coveney, 1996; Armstrong & Smith, 2002;
Hansen & Malderez, 2004) has shown that the presence or absence of ne is—at least to some
extent—determined by the speaker's social identity.
Along the social-group or interspeaker dimensions of linguistic variation, ne is
a grammatical variable of the type that responds to a variationist analysis; that
is, the variable occurs frequently enough for speakers to be able to employ in a
probabilistic way . . . the (non)standard variant to signal various aspects of
their social identity. (Armstrong & Smither, 2002, p. 28)
Specifically, age "is indeed an important variant with ne retention or deletion"
(Ashby, 1981, p. 683). In Ashby (1981) and (2001), the age group 51-64 retained ne at
relatively high rates (52% in the 1976 data and 25% in 1995), while younger speakers, ages
14 to 21, retained ne at rates of only 19% and 15% in the respective corpora. Although
overall ne retention rates had dropped in the period between the two studies, the data suggest
strongly that older speakers tend to use ne more frequently than do younger ones. Another
18
possible explanation is that the observer's paradox may be more pronounced during
interviews with older informants, and it is possible that non-retention would be more manifest
in the observer's absence. Ashby (2001) does not, unfortunately, address this issue.
Coveney (1996) reported similar results, as informants between the ages of 50 and 70
years (all women) used ne at a rate of 28.8%. Not surprisingly, the youngest informants, aged
between 17 and 24 years, had the lowest retention rates—only 8.4%. Once again, "age of
speaker emerges as the most important differentiating factor" (Coveney, 1996, pp. 86-87).
Results reported in Hansen & Malderez (2004) confirm that age is the most important social
factor concerning ne use. The oldest speakers in their corpus—aged 51 to 64 years—retained
ne at a rate of 22.3%, while informants aged between 15 and 23 years, retained ne in only
4.6% of occurrences of negation. By comparing the four studies mentioned above, we see
that ne retention rates appear to be declining, yet are consistently higher among older
speakers.1
Table 3. Age of speaker and ne retention.
Corpus Older
Speakers Younger Speakers Overall Retention
Rate Ashby (1981) 52% 19% 37% Coveney (1996) 28.8% 8.4% 18.8% Ashby (2001) 25% 15% 18% Hansen & Malderez (2004) 22.3% 4.6% 8.2% In addition to the speaker's age, other social factors—including education level and
social class—appear to influence the use of ne. Ashby (1981) and (2001), as well as Coveney
(1996) and Hansen & Malderez (2004), identify social class as an important predictor of ne
retention rates. Upper-middle class speakers tend to retain ne at higher rates than do their
middle-class and working-class counterparts. However, in Ashby (2001), social class was
found to be less influential among younger speakers than among older speakers where "la
1 Coveney (1996) and Hansen & Malderez (2004) also provide data for different age groups. These results were not included in the table in order make the comparison between Ashby's (1981) and (2001) two generation-based age groups.
19
variable (ne) reste un indicateur d'appartenance sociale" (p. 13).2 This suggests that age is, in
fact, a more important co-variant of ne than social class.
Following Ashby (2001), Hansen & Malderez (2004) also support the claim that, all
things being equal, age is the one social factor that is of most importance. Although education
level was explored and found to be somewhat influential, the authors conclude that "l'âge des
locuteurs semble structurer la variation de manière beaucoup plus convaincante que le niveau
d'études" (p. 19).3 Geographic origin was also examined; however, its influence on the
variable ne remains uncertain since those informants coming from different parts of France
normally arrived in the Paris area after the age of 25 for work-related reasons (no children
were classified as coming from outside the Paris area). Thus, age seems to be, once again, the
most important social factor.
A final social factor appears to have some influence on the variable use of ne: the
speaker's gender. Ashby (1981) reported that women were more likely to omit ne than men,
and he provides overall retention rates of 30% and 42% for women and men, respectively.
Ashby (1981) noted that "[i]f one accepts the theory that ne is indeed now being lost in
French, it appears that women are in the vanguard of this change" (p. 685); however, Ashby
(2001) found that the speaker's gender was not as important as it had been in the corpus from
1976, as men and women retained ne in 20% and 17% of all negations, respectively, in the
1995 corpus. Coveney (1996) also reported that ne retention rates were approximately
equivalent between men and women (16% and 15% respectively). Similar results were found
by Hansen & Malderez (2004), who found that the speaker's gender "ne semble pas jouer de
rôle décisif" (p. 18).4
2 Translation: "the variable (ne) remains an indicator of social class." 3 Translation: "the speaker's age seems to structure the variation much more convincingly than does his or her level of education." 4 Translation: "does not seem to play a decisive role."
20
Although these results might suggest that the speaker's gender has become less
influential over the past two and a half decades, they also support Ashby's (1981) claim that
women are leading change in progress if, of course, one accepts that men have simply
"caught up" linguistically to women. However, it is unfortunate that Ashby (2001) does not
offer any other explanations concerning the effect of the speaker's gender on ne retention in
either of his corpora. It is entirely possible, for example, that Ashby's female informants
were—for reasons unknown—more at ease with him during the interviews than their male
counterparts in the 1976 data, and, therefore, they produced fewer tokens of ne.
2.3 Phonological Factors
Ashby (1981) provided evidence that the variable use of ne is also influenced by a
certain number of phonological factors. Ashby noted that ne was "likely to be retained
postpausally . . . and in intervocalic position, provided one of the vowels is nasal" (p. 677).
However, Ashby found no evidence of regressive nasal assimilation when a consonant
precedes the ne position, which differs from what appears to occur in Montreal French
(Sankoff & Vincent, 1977).5
It is also surprising, according to Ashby (1981), that ne retention was not favored
between two nasal vowels; however, he argues that had there been more than six such tokens
in the corpus, the probability for ne retention in such a phonological environment would have
been higher. Coveney (1996), following Pohl's (1968) observation, suggested that this
phonological environment did indeed merit investigation.
Another phonological factor that affects the variable use of ne is coalescent
assimilation. For example, Coveney (1996) provides the following cases: je sais pas [ʃɛ pɑ]
and je suis pas [ʃɥi pɑ]. In these cases, not only is the schwa deleted, but [ʒ] and [s] are
5 Assimilation is the process by which a phoneme is changed to match an adjacent one, usually to facilitate pronunciation in rapid speech. In regressive nasal assimilation, a following nasal vowel influences the preceding phoneme (e.g., tu en as t'en as).
21
combined, producing [ʃ]. According to Coveney (1996), "[i]n such instances, when the
phonological environment is so radically different according to whether the ne is present or
absent, it seems more reasonable to say that the grammar is constraining the phonology,
rather than vice versa" (p. 78). Coveney even proposed that one possible solution to this
problem would be to exclude such cases from quantitative studies of ne; however, this would
eliminate a large number of tokens of negation. He decided, therefore, not to "attempt any
general quantification of phonological environment" (p. 78), but to bear it in mind, as it may
be a contributing factor in a number of syntactic and grammatical environments.
Tokens of negation found in one phonological environment have, however, been
considered impossible to study: the presence of [n] preceding a verb that begins with a vowel
(or vowel sound). Since ne [nə] becomes n' [n] in the prevocalic position, it is very
difficult—if not impossible—to distinguish prevocalic [n] of the negative particle ne from [n]
resulting from elision. For example, the sentence on n'est pas "one is not" or "we are not"6 is,
phonetically, the same as on est pas (i.e., both sentences are pronounced [õ̃̃̃nɛpɑ]). This
particular environment does not lend itself to the study of ne use in spoken French; thus,
tokens of negation in this phonological environment have been excluded from previous
studies (see Ashby, 1981 and 2001; Coveney, 1996; Armstrong & Smith, 2002; Hansen &
Malderez, 2004).
2.4 Linguistic Factors
A certain number of internal linguistic factors have also been demonstrated to
influence the variable use of ne: specifically, Neg2 type, lexicalization, grammatical subject,
and sentence type have been found to be important differentiating factors (see Ashby, 1981;
Coveney, 1996; Armstrong & Smith, 2002; Hansen & Malderez, 2004).
6 The pronoun on can be used for indefinite reference to mean "one" and for definite reference to mean "we." Peeters (2006) provides an overview of the various possible referents for the pronoun on.
22
According to Armstong & Smith (2002), more frequently occurring Neg2s—most
notably pas—co-occur less frequently with tokens of ne, confirming what Ashby (1981) and
Coveney (1996) had previously found in their respective corpora. Results reported by Hansen
& Malderez (2004), who found that "plus les éléments entrant dans l'expression négative sont
fréquents, plus il y a chute du ne" (p. 25),7 suggest that this occurs in Parisian French as well.
Table 4 shows ne retention rates according to Neg2 type and frequency as reported by Ashby
(1981), Coveney (1996), Armstrong & Smith (2002) and Hansen & Malderez (2004).
Table 4. Retention rates of ne reported in previous studies according to Neg2 type. Ashby (1981) Coveney (1996) Armstrong &
Smith (2002) Hansen &
Malderez (2004) Neg2
Tokens Retention Tokens Retention Tokens Retention Tokens Retentionpas 2,330 33% 2,317 16.4% 1,748 70.5% 941 8.2% plus 127 51% 209 25.8% 85 77.6% 96 9.4% rien 104 34% 146 21.2% 57 80.7% 61 6.6% jamais 73 36% 84 26.2% 57 80.7% 35 11.4% que 115 59% 109 34.9% 60 95% 23 30.4% personne 20 75% 24 33.3% 3 33.3% 10 0.0% aucun N/A N/A 33 21.2% 50 82% N/A N/A Multi.8 24 41% N/A N/A N/A N/A N/A N/A Table 4 shows clearly that there is a general tendency to omit ne more often when the
Neg2 occurs more frequently. However, que and plus seem to contradict this pattern. Ashby
(1981) explained that this may occur since "[t]hese second negatives may not be used
elliptically, as jamais, rien, and personne may be" (p. 679).9 It also seems that semantic
factors may play a role since "que and plus may be less categorically negative than the other
second negatives" (Ashby, 1981, p. 679). Armstrong & Smith (2002) commented on this
phenomenon as well, suggesting that ne may be used in order to avoid confusion or
ambiguity; especially in the case of plus since plus can be used as a Neg2 or in a positive
sense. However, the authors admitted that the affirmative plus is most usually pronounced
7 Translation: "the more frequent the negative elements are, the more ne is dropped." 8 "Multi." refers to instances of multiple negation; that is, two or more Neg2s are used (e.g., je n'ai plus rien). 9 For example, the question "Qui est venu ce soir?" 'Who came this evening?' can be answered elliptically by the Neg2 Personne ('Nobody').
23
with the final [s] (i.e., [plys]), whereas the Neg2 is not (i.e., [ply] or even [py]) (p. 37).
Hansen & Malderez (2004) added another Neg2 type: pas followed by an adverb (see
also Coveney, 1996). For example, pas tellement, pas vraiment, pas du tout were included in
the [pas + adverb] category, and ne retention was only 3.9%—a rather striking difference
when compared to the 8.2% retention rate reported in the pas category. Additionally, Hansen
& Malderez (2004) treated the expression pas mal as a separate category; and, of the 35
occurrences of this expression in the corpus, not one included ne. Coveney (1996) had
previously reported similar results, which suggests that a certain number of [pas + adverb]
combinations (especially pas mal and pas cher) "might almost be thought of as single lexical
items" (p. 80). Following Coveney (1996), Hansen & Malderez (2004) went even further,
suggesting that "[i]l s'agit peut-être d'une espèce de lexicalisation récente des expressions de
ce type qui feraient perdre le statut proprement négatif de pas" (p. 23).10 Although far from
conclusive, the data in the Hansen & Malderez corpus suggest that a number of [pas +
adverb] combinations are becoming lexicalized in modern French since little or no difference
was observed in the Péretz-Juillard corpus between [pas + adverb] retention rates and pas-
alone rates. If the lexicalization of other pas + adverb combinations is indeed occurring, it is
most likely following the model of pas mal (Coveney, 1996; Hansen & Malderez, 2004).
In addition to the expression pas mal and other [pas + adverb] combinations, it
appears that certain expressions are "pre-formed sequences" as opposed to novel sentence
structures. These sequences disfavor ne retention since they occur at such high frequencies
that they seem to have undergone the process of lexicalization, albeit not entirely in all cases
(Coveney, 1996). In particular, the sequences je (ne) sais (pas), je (ne) suis (pas), ce (n') est
(pas), il (n') y a (pas), and il (ne) faut (pas) have been classified as pre-formed sequences
(see Ashby, 1981; Moreau, 1986; Coveney, 1996; Armstrong & Smith, 2002; Hansen &
10 Translation: "It may be a sort of recent lexicalization of this type of expression, which would cause pas to lose its negative status."
24
Malderez, 2004). In addition, a certain number of other subject-verb sequences have also
been selected as candidates for pre-formed sequences (see Coveney, 1996; Hansen &
Malderez, 2004). Ne is frequently omitted in these cases because these sequences are used at
such high frequencies in affirmative clauses.
Since affirmative clauses are far more numerous than negative ones, it could
be argued that it is essentially the subject + verb sequence which speakers
operate with, regardless of whether a negative item follows or not. (Coveney,
1996, p. 79)
Hansen & Malderez (2004) concurred with Coveney (1996), noting that the
"hyperfréquence des verbes être, avoir, savoir et pouvoir" (p. 25, italics in original),11 which
co-occur frequently with the subject clitics je, tu, il, and ce, and the Neg2 pas, provides an
environment that strongly disfavors ne retention. Hansen & Malderez (2004) reported that ne
retention with frequently occurring verbs (i.e., verbs that occurred more than 20 times in the
corpus) was at only 7.2%, whereas less frequently occurring verbs collocated with ne at a rate
of 13.3%. Ashby (1981) and Coveney (1996) reported on ne retention in several specific
sequences, as shown in Table 5.
Table 5. Ne retention rates in preformed sequences (Ashby, 1981; Coveney, 1996). Sequence Ashby (1981)
Overall Retention = 37% Coveney (1996)
Overall Retention = 18.8% je sais 14% 8.6% je suis N/A 14.3% c'est 7% 3.6%12
il y a 24% 2.4% il faut 16% 10.5% Both Ashby (1981) and Coveney (1996) demonstrated clearly that ne retention in these
sequences was well below overall retention rates in their respective corpora. Incidentally, the
sequences il y a and il faut are somewhat problematic since the subject clitic subject il is
11 Translation: "The extremely high frequency of the verbs être (to be), avoir (to have), savoir (to know), and pouvoir (to be able)." 12 Coveney (1996) includes ça est in this category.
25
often dropped (either completely or partially)13 in spoken French (Coveney, 1996; Sax,
2003). Ashby (1981) did not mention this, and it is unclear whether or not his data included
instances of both il y a [iljɑ] or [ijɑ] and y a [jɑ]. Coveney (1996) considered both forms in his
data; however, he noted that "once the impersonal pronoun il has been omitted (as it is
extremely frequently), there is a very strong tendency to also omit ne, in order to avoid the
rare initial cluster [nj]" (p. 81). Hansen & Malderez (2004) took a different position and
excluded from their data examples in which the subject clitic had been dropped (e.g., y a pas,
faut pas) since, according to them, ne omission was, in this case, obligatory.
In addition, other types of subjects appear to influence ne use. For example, clauses in
which the subject is a noun phrase (NP) tend to favor ne retention, while those in which a
subject pronoun is used tend to favor ne omission. Table 6 shows this tendency as reported in
previous studies.
Table 6. Ne retention according to subject type as reported in previous studies. Subject
Type Ashby (1981) Coveney (1996) Armstrong &
Smith (2002) Hansen &
Malderez (2004) NP 78% 67.2% N/A 56.4% non-clitic pronoun14
57% N/A 89.2% N/A15
clitic pronoun 28% 14.6% 61.7% 5.8% Table 6 clearly demonstrates that clauses whose subject is a NP favor ne retention, as these
rates are well above the overall rate reported in all three studies that examined this factor. It is
also clear that ne retention is not favored in clauses containing a clitic subject pronoun, as
these rates are consistently well below the overall rate. Ashby (1981)—among others
(specifically Fonseca-Greber & Waugh, 2003a, 2003b)—have suggested that clitics "are now
in the process of becoming bound to the verb at the morphological level" (p. 680), which may
13 Clitic il can either be dropped completely, or the [l] may be deleted, leaving only the [i] audible (e.g, i faut [ifo] or i y a [ija]). 14 This category includes instances of cela and quelqu'un (Ashby, 1981). 15 Hansen & Malderez (2004, pp. 21-22) count non-clitic and clitic pronouns together; their results are, therefore, somewhat difficult to compare to the other studies.
26
be contributing to the loss of ne.
As the subject clitic and verb grow more and more closely bound, ne, which
can only occur between them, may be progressively squeezed out. And since it
has become only a redundant mark of negation in modern French, ne can
easily be dispensed with. (p. 681)
Coveney's (1996) data support the hypothesis that the use of a subject clitic disfavors
ne retention, yet he offered a different line of explanation.
Clitic subjects collocate with negated verbs far more frequently than do NPs,
not only because they are about ten times more frequent in discourse, but also
because they form a closed class, whereas NPs are, of course, an open, indeed,
infinite set. (p. 73)
Although ne retention appears to be disfavored when a subject clitic is used, ne retention rates
appear to vary between the various subject clitics. More specifically, je, tu, on, and ce have
been found to collocate with ne at a very low frequency, regardless of the Neg2 and verb used
in the negation. On the other hand, il, elle, and ils collocate with ne at a relatively higher rate,
(see, for example, Hansen & Malderez, 2004). Extremely low rates of ne retention appear to
be, according to Ashby (1981), a result of another continuing change in modern French: "the
fusion of the clitic pronoun and the verb" (p. 868).
Another linguistic factor that appears to influence the variable ne is clause type; and,
by extension, verb tense and form (e.g., simple or compound, personal or impersonal, etc.).
Ashby (1981) reported that ne was retained 31% of the time in declarative clauses, while ne
was retained categorically in imperative clauses. Armstrong & Smith (2002) reported a high
rate of ne retention in imperative clauses as well (95%).
Along the same lines, it has been demonstrated that ne retention remains relatively
high in co-occurrence with negated impersonal verb forms (i.e., those that are not conjugated
27
with a subject, such as present participles and infinitives). Ashby (1981) reported a ne
retention rate of 68% in this environment and Hansen & Malderez (2004) reported a rate of
26.3%. Coveney (1996) and Armstrong & Smith (2002) separated infinitives and present
participles and reported the following retention rates.
Table 7. Ne retention in [- overt] subject environments as reported by Coveney (1996) and Armstrong & Smith (2002).
Corpus Retention with pres. part. Retention with infinitives Coveney (1996) 50% 62.5% Armstrong & Smith (2002) 66.7% 93.6% Armstrong & Smith (2002) offered one possible explanation for higher rates of ne retention
in this particular environment.
It seems likely that the relative rarity of negated verbs lacking overt subjects
leads the speakers to focus on the negative markers in the utterance more than
they would do in expressions with subjects, and this greater self-monitoring
would naturally entail a higher rate of ne retention. (p. 36)
Other verb forms have also been taken into consideration; in particular, retention rates
in compound and simple tenses have been compared. Ashby (1981) provided evidence that ne
retention was more likely in compound tenses, as retention rates with the auxiliary verbs être
and avoir were 55% and 50%, respectively, whereas ne was retained 35% of the time with
lexical verbs. Hansen & Malderez (2004) reported similar findings, as retention rates in
compound tenses reached 13.2%, while ne was retained only 7.8% of the time in simple
tenses. These results suggest that compound tenses favor ne retention. Ashby (1981) also
found evidence that ne was less likely to be retained with the aspectual auxiliary aller (29%
of all instances).
2.5 Stylistic Factors
It has been frequently suggested in the literature that ne is dependent on a number of
stylistic factors; primarily formality and discourse topic. The influence of formality and
28
discourse topic is particularly evident when one compares results reported in Armstrong &
Smith (2002) with those reported in studies concerning more informal speech (Ashby, 1981;
Coveney, 1996; Hansen & Malderez, 2004).
Ashby (1981) attempted to quantify this factor by comparing ne retention in the first
half of each interview with informants with that of the second half. He asserted that
interviewees became more at ease with him as the conversation progressed, and, as a result,
their speech became less guarded in the later half of the conversation. Ne retention should,
therefore, be lower in the second half of the interview than in the first half. This was shown to
be true—at least to some extent—as retention rates dropped on average by 2% (from 37% to
35%). Although these results suggested that speakers retain ne at higher frequencies when
they believe that the context requires a more formal or more highly monitored speech style,
they are less than conclusive. However, Ashby (1981) was able to follow three informants as
they "moved between two widely different social settings" (p. 681). Ne retention rates for
these three speakers dropped from an average of 37% in the formal setting to only 16% in the
informal setting, thereby confirming Ashby's presumption.
It has also been suggested that discourse topic may have an effect on the variable use
of ne. Indeed, Ashby (1981) explored this aspect by selecting five speakers who had the
lowest rates of ne retention. He identified 24 tokens of ne and attempted to determine why
these tokens were produced by these speakers. Four ne tokens occurred when the speaker was
responding to or repeating a question in which ne had been used. Another two tokens
occurred after a false start, which suggested self-monitoring on the part of the speaker. One
token was used in a direct quotation and five were produced while talking about disciplining
children. Another five tokens were classified as proverbial, following the example of Sankoff
& Vincent (1977),16 and two more were produced during the discussion of serious topics.
16 Proverbial refers to truisms or general statements or rules (see Sankoff & Vincent, 1977).
29
Only four tokens, then, seemed to escape explanation, but Ashby did not comment on any
other linguistic or phonological factors that might have led to their presence. Results reported
in Armstrong & Smith (2002) provide the most conclusive evidence that ne retention is
strongly favored in highly monitored speech (i.e., formal contexts).
Coveney (1996) found that one informant in particular had ne retention rates that
differed greatly from the scores of the others. In his opinion, this was due—at least in part—
to the fact that this young male was an assistant camp director and may have felt the need to
represent the camp favorably by speaking more formally. Further, the informant asked during
the interview if he should speak slowly in order to make his speech more easily understood
by native speakers of English, which presumably resulted in a more formal register of speech
(i.e., less similar to his vernacular since more attention was paid to his speech). The speaker's
belief that a more formal register of speech was required led Coveney (1996) to consider his
scores separately; however, this supported the hypothesis that speakers retain ne at higher
frequencies in situations that, in their view, require more a formal register of speech; for
example, sociolinguistic interviews with a researcher (Labov, 1972).
Ashby (2001) acknowledged the effect that his presence—as well as the presence of
the tape recorder—might have had on informants, yet insists that speakers in his study were
at ease during the interviews.
Ces conditions n'ont certes pas favorisé un registre tout à fait familier, où la
chute du ne serait probablement encore plus manifeste, mais dans l'ensemble,
au bout de quelques minutes les locuteurs avaient l'air assez à l'aise, comme
s'ils oubliaient la présence du magnétophone, et ne manifestaient aucune
réticence à parler. (p. 8)17
17 Translation: "These conditions certainly did not favor a completely familiar register of speech, where ne omission would probably be more manifest, but as a whole, after a few minutes the speakers seemed at ease, as if they had forgotten about the tape recorder, and showed no reluctance to speak."
30
As Coveney (1996) and Ashby (2001) pointed out, sociolinguistic interviews may be
perceived as a more formal speech environment by some speakers; much like the radio
interviews analyzed by Armstrong & Smith (2002), albeit to a lesser extent.
Hansen & Malderez (2004) brought another interesting stylistic factor into
perspective: ne production in children's written school assignments. Since children between
the ages of 5 and 14 years had omitted ne categorically in their corpus of informal speech, the
authors decided to explore ne use in written school assignments of young children. Hansen &
Malderez (2004) reported that ne retention in these written assignments was as high as 75%
with noun subjects and 50% with pronominal subjects—a striking difference between rates
reported in the speech of similarly aged informants. Likewise, Blanche-Benveniste (1997)
noted that children in her corpus were perfectly capable of using ne appropriately when
imitating the speech of elegant women. These results suggest strongly that ne is not being
completely lost from the French language. Hansen & Malderez (2004) posited that "on est
encore très loin d'un tel aboutissement pour la simple raison que les locuteurs se trouvent
exposés à l'usage de ce ne dans une diversité de contextes qui le retiennent certainement dans
leur système linguistique; d'abord dans le français parlé formel" (p. 26)18
2.6 Summary of the Relevant Literature
Previous studies clearly indicate that ne use is indeed dependent on a number of
social, linguistic, and stylistic factors. Age, in particular, stands out as an important
demographic factor, as does the social class to which the speaker belongs. It is also clear that
ne use is a sociolinguistic variable that might be considered a marker of social identity since
retention rates tend to be higher among older speakers of the higher social class (Ashby,
2001).
18 Translation: "we are still far from such an outcome for the simple reason that speakers are exposed to the use of this ne in diverse contexts, which retains it in their linguistic system; firstly, in formal spoken French."
31
Among the phonological and linguistic factors, subject type might be considered a
determining factor, especially when a subject clitic is used and assimilation is possible (see
Ashby, 1981; Coveney, 1996). Other important factors include verb type and form, Neg2
type, and frequency of [subject + verb] sequences.
Style remains a determining factor since highly monitored speech styles appear to
favor ne retention (Armstrong & Smith, 2002). It is also important to remember that the
speaker's perception of the required level of formality of a given communicative context
plays a role (Coveney, 1996). Finally, as Hansen & Malderez (2004) noted, even young
children who omitted ne categorically in informal speech were able to use it appropriately in
written assignments. In addition, children appear to be capable of using 2Neg in imitation of
certain stylistically marked contexts, such as the speech of elegant women (Blanche-
Beneveniste, 1997).
The present study attempts to address many of the same factors explored in the
aforementioned research. Although interpersonal variation and demographic factors have
always been taken into consideration in previous variationist studies, the present study does
not attempt any such quantification since informants in chat environments are able to remain
more or less anonymous, and it is impossible to gather (credible) personal data about such
informants. However, as mentioned in 1.3.4, the anonymity granted by the chat environment
provides an ideal context in which to observe informants' vernacular (at least as far as
synchronous CMC is concerned). It is reasonable to assume that the informants' language is
less guarded than that of informants in sociolinguistic interviews since it is less directly
observable (i.e., informants in chat environments do not participate in interviews with a
researcher and are more or less unaware that observation is taking place).
32
CHAPTER 3
METHODOLOGY
3.1 The IRC Environment
Internet Relay Chat (IRC) is a protocol that enables users to chat in real time with one
or more interlocutors by sending text-based messages when logged on to a specific chat room
(i.e., chat channel). Chat rooms may be public or private and are generally hosted by a
network, which is itself connected to a server. IRC servers are freely accessible, provided that
the user has access to an Internet connection and has downloaded one of several IRC clients
available on the Internet.1 The IRC environment is rather anonymous, and participants see
only the pseudonyms (i.e., screen names) of their interlocutors.
Once logged on to a channel, the user simply types a message and presses the "enter"
button to send it. Any person connected to the chat channel will see the message. Since
communication in the IRC environment occurs in real-time and several (sometimes dozens)
of participants are often sending and responding to messages simultaneously, chat discussions
can appear to be random and are potentially difficult to follow for the non-initiated. Excerpt 1
provides an example of data taken during an IRC discussion.
Excerpt 1. <Ayame> jme suis acheté une épée XD <Okko> Qu'est-ce que j'ai 'core fait ? <KaM> bisouuuuuuuuuuuus ma tite Ayame au fait :p <Okko> Ayame oO <Okko> pourquoi faire ? couper les carottes ? :o <romanticboy> lu Ayame <romanticboy> tu va bien? <romanticboy> lu Okko
1 An IRC client is a program which connects a user to his or her choice of IRC servers. Pierozak (2003c) provides a detailed description of the French-language IRC environment.
33
<Okko> salut romanticboy * taku`[Oforum] is now known as taku <romanticboy> j t ai deja di hier Ayame ca te monte a la tete ces jeux! <romanticboy> lolllllllllllll <romanticboy> ;) <Ayame> ça va et toi romanticboy ? <luka> moi aussi g une épée :) <Ayame> Okko pour décorer ^^ <Ayame> mecmsn20 dégage <Okko> Tu sais Ayame, pour le self-defense, y a mieux que les épées, c'est pas très discret pour sortir dans la rue, surtout en cette saison, j'te vois mal avec un pardessus pour cacher le bazar :o ['<Ayame> i bought myself a sword XD <Okko> What have i done now? <KaM> mwaaaaaaa my little Ayame actually :p <Okko> Ayame oO <Okko> to do what? cut carrots? :o <romanticboy> hey Ayame <romanticboy> how're you? <romanticboy> hey Okko <Okko> hey romanticboy * taku`[Oforum] is now known as taku <romanticboy> i already told you yesterday Ayame these games are going to you head! <romanticboy> lolllllllllllll <romanticboy> ;) <Ayame> fine and you romanticboy ? <luka> i also have a sword :) <Ayame> Okko for decoration ^^ <Ayame> mecmsn20 get out of here <Okko> You know Ayam, there's better than swords for self-defense, it's not very discreet for going out, especially at this time of year, I have trouble picturing you in a trench coat to hide it :o'] As excerpt 1 illustrates, several discussions are taking place at the same time, and in
the space of approximately two minutes, six different users send messages. As Herring
(1999) noted, communication in the chat environment can be somewhat incoherent. Despite
the apparent lack of interactional coherency, however, IRC and other forms of synchronous
CMC have become popular communicative environments around the globe.
34
3.2 Data Collection
The data used in this study were collected from two different age-based chat channels
(#18-25ans and #25-35ans) found on the public IRC server EpikNet.2 Data was collected
over the course of four days—selected at random—for approximately four hours each time
during the fall of 2005. The chat discussions were recorded using the transcript recording
function provided in the mIRC software3 and saved as text files for analysis. The corpus
constitutes a wide range of conversation topics and discussions held during different times of
day (afternoon and evening), as well as during both weekday and weekend times.
During data collection times, I did not actively engage in the discussion, nor did I
reveal my identity as a researcher. The chat participants were—as far as can be known—
unaware that observation was taking place.4 However, it must be noted that since this type of
communication occurs in a public space, the participants are certainly aware of the possibility
that any number of people could be following the chat session or reading the log.
Following data collection, the transcripts of data were reviewed and analyzed, and all
server- and human-generated turns were counted.5 In order to determine the size of the
corpus, every human-generated word was counted using a concordance program,6 which will
be discussed in the following section.
Table 8 shows the number of turns (both server- and human-generated) and the
number of human-generated words found in the corpus.
2 EpikNet is a free-access text-based IRC server with a large selection of French language chat channels. More information about EpikNet can be found at the following URL: http://www.epiknet.org/ 3 mIRC software is one of several programs that allows IRC users to access different IRC severs. Information about mIRC software can be found at the following URL: http://www.mirc.com/ 4 This project—including data collection method—was reviewed and approved by the Institutional Review Board for the Protection of Human Subjects of the University of North Texas. Project application #05-357. 5 I have opted to define a turn as a message sent by a participant when he or she hits the Enter key (human-generated) or when the server sends a message (server-generated). This definition is not, however, unproblematic. For a discussion of turn-taking in chat, see Thorne (1999, ch. 5) or Williams (2003, ch. 3). For a general treatment of turns, see van Lier (1988). 6 Information about Concordance© software can be found at the following URL: http://www.concordancesoftware.co.uk. A detailed explanation is provided in section 3.3.
35
Table 8. Turns and Human-generated Words Chat
Channel Total No. of
Turns Human
Gen. Turns Server
Gen. Turns Human Gen.
Words #18-25ans 8,774 (47.78%) 6,864 (78.23%) 1,910 (21.77%) 37,752 (48.32%) #25-35ans 9,591 (52.22%) 7,690 (80.18%) 1,901 (19.82%) 40,373 (51.68%)
Total 18,365 (100%) 14,554 (79.25%) 3,811 (20.75%) 78,125 (100%) Server-generated turns were not analyzed further since the objective of this study is to
investigate the variable use of ne by human participants engaged in chat discussions. The
human-generated turns were reviewed and analyzed, and every token (i.e., occurrence) of
negation was identified and coded as described in the following sections.
3.3 The Concordance© Software and Counting Methodology
Concordance© is a program that allows the user to compile a complete concordance
of all words found in a text document, the frequency of each word, and where it can be found
in the text. Figure 1 is a screen capture of the Concordance© software.
Figure 1. Concordance© software: headword list and occurrences.
A list of headwords found in the file is shown on the left side of the screen and the
number of occurrences is provided. By selecting one of the headwords from the list, the user
is able to view all occurrences of that word to the right. In Figure 1, the word pas has been
36
selected and all occurrences of pas found in the text are shown to the right, as well as where it
appears in the text.
In addition to providing a list of headwords and their location in the text file,
Concordance© allows the user to view each occurrence in its context by selecting one of the
examples found in the right window. A separate window appears in which the selected
occurrence is highlighted, as shown in Figure 2.
Figure 2. Concordance© software: text viewer. The text viewing function is especially useful when reviewing excerpts of data since
the user is able to view the token in its context (i.e., in order to read turns preceding and
following the excerpt).
Although Concordance© is very efficient, certain problems did arise; specifically,
typographical errors and abbreviations commonly found in chat discourse, as well as
conjunctions with punctuation, were counted inconsistently by the program. This
inconsistency often resulted in two or more words being counted as one (e.g., pask'elle or
parce qu'elle was counted as one word), among other anomalies. In addition, ne was often
omitted from the head-word list because of punctuation or spacing errors (e.g., jene sais pas
37
would have been counted as three words: jene, sais, pas).
In light of these problems, the head word list and many examples were carefully
reviewed in order to ensure an accurate word count. All occurrences of Neg2s were also
carefully reviewed and every example of negation was counted and coded as described in
section 3.4.
3.4 Examples of Negation Found in the Corpus
Every token of negation in the corpus was identified and classified as either two-
particle negation (2Neg) or single-particle negation (1Neg). Tokens of 2Neg were reviewed
and instances of fixed expressions (e.g., n'est-ce pas, ne serait-ce pas, etc.) were eliminated.7
Instances of 1Neg were then classified as one of six types shown in Table 9.
Table 9. Types of 1Neg found in the corpus. Type Description A Non-verbal sentence B Non-verbal sentence preceded by a verbal clause C Fixed expressions involving a Neg2 D Verbal clause (non-fixed expression) E Verbal clause in which a clitic pronoun has been deleted F Non-traditional form
The following excerpts of data illustrate the different 1Neg types found in the corpus.
The negation of interest has been underlined. The term sic has not been used to indicate that a
grammatical, orthographic, or typographic error has merely been reproduced. A literal
English translation is provided in brackets immediately below each excerpt. A second
translation in more "natural" English is also provided (in single quotation marks).
1Neg Type A is usually a short response to a question or statement sent by another
participant, as illustrated by Excerpt 2. This type of 1Neg can also be a question or statement
by itself, especially when followed by an adjective, adverb or past participle (i.e., the subject
and verb are not present). Excerpt 3 provides an example of this use.
7 Hansen & Malderez (2004) provide this methodology since these expressions require ne to be present.
38
Excerpt 2. <Eliot> pas vraiment non [<Eliot> [Neg2] really no] '<Eliot> not really no' Excerpt 3. <Eliot> pas trop oqp ange_away? [<Eliot> [Neg2] too busy ange_away?] '<Eliot> not too busy ange_away?' 1Neg Type B is a non-verbal clause, which is similar to Type A. However, Type B is
immediately preceded by a verbal clause to which the Type B negation is normally linked, as
shown in Excerpt 4.
Excerpt 4. <Prue> n'empêche que c'est pas forcément vrai, mais pas faux non plus Hugo :) [<Prue> [Neg1]-stop8 that it's-[Neg2] necessarily true, but [Neg2] false either Hugo :)] '<Prue> albeit it's not necessarily true, but not false either Hugo :)' 1Neg Type C includes instances of the expression pas mal. This expression is
generally considered to have been lexicalized without ne in modern French (see Coveney,
1996), although there is some debate (see Hansen & Malderez, 2004). Excerpt 5 provides an
example of pas mal use in a verbal clause. In addition, this expression can also be used as a
quantifying adverb in lieu of beaucoup as shown in Excerpt 6. For these reasons, instances of
pas mal have been counted separately from other instances of 1Neg. A number of examples
of the expression pas cher have also been included in this category.9
Excerpt 5. <romanticboy> c pa mal [<romanticboy> it's [Neg2]-bad] '<romanticboy> it's not bad' Excerpt 6. <ToUfOu> j'ai pas mal bossé [<ToUfOu> i've [Neg2]-bad worked] '<ToUfOu> i worked alot' 8 Neg1 refers to the preverbal ne. The expression n'empêche que always includes ne (Grevisse, 1993). Such examples have not been counted as tokens of the negative particle ne. 9 Most instances of pas cher have been counted as 1Neg Type D. However, it has been argued that pas cher is becoming lexicalized in French (see Coveney, 1996) since it can be used in an expression such as on peut trouver des voitures à pas cher. Such examples of pas cher were counted as Type C.
39
Type D of 1Neg is the most often occurring type, which includes a negated verb form
as well as a subject (e.g., a noun phrase or pronoun), except in the case of imperatives and
impersonal moods. Examples of 1Neg Type D constitute instances of ne deletion since the
presence of ne was possible (i.e., between the subject and the verb), but ne was not used (i.e.,
omitted), as shown in Excerpt 7. Thus, I have opted to define 1Neg Type D as the variant of
2Neg. In other words, the paradigm [ne present vs. ne absent] could be re-written [2Neg vs.
1Neg Type D].
Excerpt 7. <KaM> les gens vulgaire j'aime pas moi [<KaM> the people vulgar i like-[Neg2] me] '<KaM> I don't like vulgar people' 1Neg Type E occurs when chat participants attempt to imitate the spoken form of a
certain number of expressions; specifically, il faut (Excerpt 8) and il y a (Excerpt 9). As in
spoken French, the indefinite clitic pronoun il is very often deleted in chat. Other examples of
subject clitic deletion were also counted as Type E (Excerpt 10). This is another example of
how participants in this type of communication environment attempt to imitate certain traits
common in spoken discourse.
Excerpt 8. <Prue> y a pas de fontaine [<Prue> there have-[Neg2] any fountain] '<Prue> there are no fountains' Excerpt 9. <Salizar> Ben faut pas Petite-Peste xD [<Salizar> [DM]10 better-[Neg2] Petite-Peste xD '<Salizar> Well better not Petite-Peste xD'] Excerpt 10. <Devotion> ah non suis pas gentil moi :( [<Devotion> ah no am-[Neg2] nice me :( '<Devotion> ah no i'm not nice :(']
10 Discourse Marker
40
Although it seems that ne cannot be present in 1Neg Type E, these instances have not
simply been discarded. I have opted to analyze separately 1Neg Type E and to consider such
examples as indicative of the relationship between IRC discourse and everyday
conversational speech. In addition, I have included 1Neg Type E in Chapter 5 as part of the
discussion of features of discourse particular to chat.
1Neg Type F includes instances of non-traditional verb negation (i.e., grammatically
incorrect usage); specifically, imperative clauses in which the Neg2 precedes the verb, which
imitates certain spoken forms (e.g., a command for a family pet or other very informal
contexts).
Excerpt 11. <KaM> puis pas touche a ma ange_away [<KaM> [DM] [Neg2]-touch to my ange_away '<KaM> [DM] no touching my ange_away']
3.5 Coding of Tokens
After identifying every occurrence of 2Neg and 1Neg Type D, each token was coded
for analysis. GoldVarb 2001 was used in order to analyze a number of linguistic (syntactic)
factors: Neg2 type, subject type, phonological environment, and sentence type. GoldVarb
2001 is a multivariate analysis program that allows the user to test the influence of a number
of independent variables on the dependent variable (in this case ne-present vs. ne-absent) and
to determine the significance of the influence of each independent variable. Thus, it was
possible to analyze not only the variable use of ne, but also other variables that co-occur
and/or co-vary with ne. In Chapter 4, I describe GoldVarb coding in more detail during the
discussion of each factor. I have also provided a detailed coding chart in Appendix A.
Following the analysis of syntactic factors, it became clear that ne use was determined
to a large extent by discourse style (e.g., ludic, emphatic, proverbial, and explanatory). In
order to test this hypothesis, a second GoldVarb analysis was conducted for a sub-sample of
the data. I have provided more detailed information about this sample in 4.3 and the coding in
41
Appendix D.
Each token of negation was analyzed in its context. Turns preceding and following the
example in question were reviewed in order to determine which pragmatic function (e.g.,
humor, emphasis, etc.) the token fulfilled. The text-viewing function of Concordance©
software proved useful for the analysis of these tokens since it was often necessary to read
several (sometimes many) turns above and/or below the example in order to understand the
context of the conversation, which was not always immediately obvious given the
interactionally incoherent nature of synchronous CMC (Herring, 1999). It was, in fact, often
the presence of punctuation, smileys,11 capitalization, a shift in the use of pronouns, or the
reaction of another participant in a subsequent turn that made classification possible.
11 "Smileys" are a type of emoticon. The Oxford English Dictionary defines an emoticon as "a representation of a facial expression formed by a short sequence of keyboard characters (usually to be viewed sideways) and used in electronic mail, etc., to convey the sender's feelings or intended tone." For a general treatment of smileys in French-language chat, see Pierozak 2003c.
42
CHAPTER 4
ANALYSIS OF THE CORPUS
In the following sections, I examine a limited number of linguistic factors in order to
draw a comparison of the use of ne in speech and IRC discourse. In addition to exploring a
variety of syntactic and phonological environments, I present and analyze a number of
pragmatic features of discourse that appear to influence strongly the use of ne in this
communicative context.
I am operating under the assumption that IRC constitutes a more or less informal
communication environment (i.e., similar to everyday conversational speech). Moreover, the
language observed in this environment should be relatively unguarded since the participants
were unaware that observation was taking place.1 It can therefore be assumed that the data
collected and analyzed in this chapter represent the "vernacular" of the IRC environment.
4.1 Distribution of 2Neg and 1Neg
Before proceeding to the analysis of linguistic and pragmatic factors that influence ne
use, let us first consider the overall distribution of 2Neg and 1Neg in the corpus. Table 10
demonstrates clearly that most second-negatives did not co-occur with tokens of ne.
Table 10. Distribution of 2Neg and 1Neg Total Negation 2Neg 1Neg 1,648 (100%) 195 (11.83 %) 1,468 (88.17%)
Instances of 1Neg were divided into several types, depending on the nature of the
negation (see 3.4). This was done in order to determine which examples could be compared 1 It must be noted that the language used in this communicative environment is somewhat guarded since there are a number of operators, also known as "ops", who enforce the rules and regulations described in the netiquette. Profanity and the use of capital letters, for example, are prohibited and ops have the power to kick-out and/or ban any participant who does not respect these rules. This does not appear, however, to elicit a less-than-informal speech style where non-traditional forms and structures might be perceived as inappropriate.
43
with instances of 2Neg (i.e., examples of the variant of 2Neg). Table 11 divides tokens of
1Neg into six possible types of 1Neg.2
Table 11. 1Neg distribution. 1Neg
Type A 1Neg Type
B 1Neg Type
C 1Neg Type
D 1Neg Type
E 1Neg Type
F Total
247 (17.00%)
50 (3.44%) 22 (1.51%) 1,017 (69.99%)
107 (7.36%)
10 (0.69%) 1,453 (100%)
Table 11 shows clearly that 1Neg Type D is the most often occurring type of 1Neg,
representing approximately 70% of all 1Neg tokens. In the remainder of the present study, the
distribution of 2Neg and 1Neg Type D (i.e., ne present vs. ne absent) will be explored. Table
12 illustrates the overall distribution of 1Neg Type D and 2Neg.
Table 12. Overall distribution of 1Neg Type D and 2Neg. 2Neg 1Neg Type D Total
195 (16.09%) 1,017 (83.91%) 1,212 (100%) It is clear that ne retention rates are low in this corpus (16.09% overall), which
corroborates the findings of Coveney (1996), Ashby (2001), and Hansen & Malderez (2004).
This in itself seems to support the assumption that, at least as far as the variable ne is
concerned, the discourse of French-language IRC is similar to that of everyday
conversational speech.3 It is therefore reasonable to assume that IRC provides a
communication environment in which informal language and non-traditional structures are
perceived as appropriate (or at least not inappropriate).4
2 Type A = non-verbal sentence; Type B = non-verbal sentence preceded by a verbal clause; Type C = fixed expressions; Type D = verbal sentence (non-fixed expression); Type E = verbal sentence, subject omitted; Type F = non-traditional structure. See 3.4 for a full description and examples of each 1Neg type. 3 Armstrong & Smith (2002) provided evidence that, although ne deletion seems to be spreading to highly-monitored speech styles, high rates of ne retention are often associated with formal discourse, such as radio programs. Hansen & Malderez (2004) also reminded us that children, who often have the lowest retention rates, are capable of using ne correctly in written school assignments. 4 Williams & van Compernolle (2007) demonstrated, for example, that the informal second-person pronoun tu is overwhelmingly preferred to the more formal vous-singular in IRC environments, and that vous-singular use is often perceived as strange or socially inappropriate.
44
4.2 GoldVarb 2001 and Statistical Procedures
A certain number of internal linguistic factors appear to structure the variable use of
the negative particle ne in everyday conversational speech (Ashby, 1981; Coveney, 1996;
Hansen & Malderez, 2004). In section 4.2, I explore a limited number of these factors (e.g.,
second-negative type (Neg2 type), subject type, and preceding and following phonological
environment) in a corpus of synchronous French-language CMC. The data were analyzed
with GoldVarb 2001 in order to demonstrate test for the statistical significance of each factor
group.
Analyzing data with GoldVarb 2001 requires three basic steps. First, all tokens of the
dependent variable must be coded according to the factor groups established by the
researcher. For the purposes of this study, each occurrence of 1Neg Type D and 2Neg was
coded by Neg2 type, subject type, preceding and following phonological environment, and
sentence type. The GoldVarb 2001 coding-key for the analysis of the first coding is provided
in Appendix A.
In the second step, a one-level binomial analysis is performed by the program.
GoldVarb 2001 calculates the weight of each factor, or, in other words, the "probability of the
dependent variable occurring in the context" (Tagliamonte, 2006, p. 220). A GoldVarb 2001
probability score equal to or greater than .500 indicates that the dependent variable is favored,
while a score equal to or less than .499 indicates that the dependent variable is disfavored.
GoldVarb 2001 reports these scores based on the order of the dependent variables when
doing the analysis. In the present study, the analysis was performed on the bias of the
presence of ne. Therefore, GoldVarb 2001 scores reported in the following analysis state the
probability that ne will be present (i.e., ≥ .500 = 2Neg favored, ≤ .499 = 2Neg disfavored)
according to each factor group. (The one-level binomial analysis report for the first coding is
provided in Appendix B.)
45
The third and final step of a GoldVarb 2001 analysis involves what is called a "step-
up/step-down analysis," which determines the "best fit of the model to the data"
(Tagliamonte, 2006, p. 228). It is at this point that GoldVarb 2001 calculates the significance
of each factor group considered and provides additional evidence of overlapping and
interaction of factor groups. (The step-up/step-down analysis report for the first coding is
provided in Appendix C.)
In addition to GoldVarb 2001 probability scores, chi-square was used where I had not
coded for the multivariate analysis since "the chi-square test enables us to compare the
frequencies we actually observe with those we should expect to observe on the basis of some
theoretical model" (Butler, 1985, p. 112). For example, GoldVarb 2001 had been used to
analyze the distribution of one- and two-particle negation according to subject type (e.g.,
noun phrase, pronoun, [- overt] subject); however, I wished to examine the variation within
the pronoun category since much could be learned by exploring the distribution of negation
according to pronoun type (e.g., clitic vs. non-clitic). Therefore, using chi-square, I was able
to determine the statistical significance of a certain number of sub-factor groups that were not
necessarily represented in the GoldVarb 2001 analysis.5 Although chi-square has been used
in previous studies of negation in French (e.g., Hansen & Malderez, 2004), it is used
sparingly in this thesis only in order to understand linguistic variation within the system as a
whole since chi-square cannot account for intraspeaker variation or weigh the frequency of
each speaker's contribution of different types of tokens.6
5 In the remainder of this paper, any mention of chi-square should be understood as a traditional or Pearson's chi-square test. 6 For an overview of the assumptions underlying chi-square, see Hatch & Lazaraton (1991, ch. 14), Butler (1985, ch. 9), and Wilcox (1996, pp. 82-85).
46
4.3 Linguistic Factors
4.3.1 Analysis of First Coding
All tokens of 2Neg and Type D of 1Neg were coded for a variety of internal linguistic
factors, including Neg2 type, subject type, preceding phonological environment, following
phonological environment, and sentence type. After reviewing the data, it was obvious that
too few tokens of Neg2s other than pas were available for independent statistical analysis.
Therefore, all Neg2s other than pas were collapsed into one category, thereby eliminating a
number of empty cells and extremely low cell counts. In addition, subject type was divided
into three categories: noun phrase (NP), pronoun, and [- overt] subject (i.e., imperatives).
Negated infinitives were not considered in the final instantiation of the analysis for this study
due to a number of methodological problems, namely a lack of examples.
Most infinitives that were preceded by a negative particle followed a modal auxiliary,
which is an environment where the presence of ne can alter the semantic interpretation of the
negation, yet its absence leaves the phrase rather ambiguous. For example, on ne peut pas
aimer ce film ('one [Neg1]-can-[Neg2] like this film') conveys a different message than on peut ne
pas aimer ce film ('one can [Neg1]-[Neg2]-like this film'). While the former expresses an
impossibility (i.e., it is impossible to like this film) by concentrating the negation on the
modal auxiliary pouvoir, the latter expresses a possibility (i.e., it is possible to not like this
film) by focusing the negation on the following infinitive aimer. Yet when ne is absent (i.e.,
on peut pas aimer ce film), the possible nuance illustrated above disappears. In light of this
limitation, it was assumed that the modal auxiliary was negated, not the following infinitive.
After recoding the data accordingly (i.e., excluding negated infinitives), a one-step
binomial analysis was run in order to test each of the factor groups independently of one
another (see Appendix B). The input probability score of 0.143 for this analysis indicated that
the chances are 14.3% that "any given token" (Young & Bayley, 1996, p. 270) will retain ne.
47
"The input probability is the likelihood that the [2Neg] rule will operate in any circumstance,
irrespective of conditioning factors" (p. 270). The data shown in Table 13 provide factor
weights for all factor groups from the one-step binomial analysis of the first coding.
Table 13. One-level binomial: first coding. Factor Group Weight Neg2 Type
pas .492 Neg2 other than pas .530
Subject Type NP .892 Pronoun .454 [- overt] subject .511
Preceding Phonological Environment Vowel .447 Consonant .486 Nasal Vowel .544 No Preceding Phonological Environment .863
Following Phonological Environment Vowel .468 Consonant .524
Sentence Type Declarative .543 Interrogative .446 Imperative .172
The one-step binomial report revealed a number of problems in the distribution of the
data. First, no convergence was found after 20 iterations, which indicates that the data do not
fit within the expected theoretical model. Second, relatively high error scores were present in
a number of cells. According to Preston (1996, p. 11), an error score of equal to or less than
2.0 indicates a good fit. In this first coding of the data, no less than eight cells had error
scores greater than 2.0. Third, the total chi-square calculated was over 48.19, and in order for
the results to pass the goodness-of-fit test (Young & Bayley, 1996, pp. 272-273), the total
chi-square produced by Goldvarb 2001, the total chi-square (p = .05 with 9 degrees of
freedom) should have been less than 16.919. These problems suggest that two or more factor
groups may be interacting. A cross-tab analysis of factors groups 2 and 3 (i.e., subject type
and preceding phonological environment) showed very unbalanced distribution in some
48
coding strings, which also suggests interaction between factor groups.
The step-up/step-down analysis of the first coding (see Appendix C) was reviewed in
order to determine whether factor weight (i.e., constraint) rankings change as factor groups
are run against each other. The step-up/step-down report revealed that whenever factor
groups 2 and 3 (subject type and preceding phonological environment, respectively) were co-
present during the regression analysis, no convergence was reached, which strongly suggests
that the two factor groups are inextricably linked. Further, factor weight rankings for subject
type actually changed in a number of runs. As Table 13 indicates, the constraint ranking
should be, in order from ne most likely to least likely: NP, [- overt] subject, pronoun.
However, in four runs where both subject type and preceding phonological environment are
present (#8, #16, #20, and #23; Appendix C), pronouns and [- overt] subject environments are
reversed. Subject type and preceding phonological environment are clearly interacting in this
study. It is therefore necessary to separate the two interacting factor groups and run two
separate analyses, a remedy suggested by Tagliamonte (2006, p. 234).
4.3.2 Analysis of Second Coding: Phonological Environment
Preceding and following phonological environment were first examined
independently of all other factor groups in the second coding. The one-step binomial analysis
for phonological environment (see Appendix D) indicated as 15.3% the probability that the
rule (i.e., the presence of ne) would be applied to any given token, regardless of the
environment. In addition, the total chi-square was below the value required (p = .05, 4
degrees of freedom), indicating that the data are a good fit to the model. In addition, only one
cell had an error score greater than 2.0; however, Tagliamonte (2006, p. 221) argues that this
is not uncommon in a distributional analysis such as this. Table 14 provides the factors
weights reported by GoldVarb 2001 in the one level analysis.
49
Table 14. One-level analysis of phonological environment factor weights: second coding. Factor Group Weight Preceding Phon. Environ.
Vowel .445 Consonant .712 Nasal .604 No preceding phon. environ. .594
Following Phon. Environ. Vowel .484 Consonant .512
Given the apparent good fit of the distribution of 2Neg and Type D of 1Neg, the step-
up/step-down analysis was performed for phonological environment (see Appendix E).
GoldVarb 2001 found preceding phonological environment to be significant. Table 15
provides the distribution of 2Neg and Type D of 1Neg according to this factor.
Table 15. 2Neg and 1Neg distribution according to preceding phonological environment. Environment 2Neg 1Neg Type D Total Varbrul Score Vowel 114 (12.67%) 786 (87.33%) 900 (100%) .444 Consonant 45 (30.82%) 101 (69.18%) 146 (100%) .710 Nasal vowel 14 (21.88%) 50 (78.12%) 64 (100%) .607 No preceding phon. environ.
22 (21.57%) 80 (78.43%) 102 (100%) .602
Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%) —
The data in Table 15 indicate that the retention of ne is disfavored when the
phonological environment immediately preceding the ne position is a vowel (or vowel
sound). However, when a consonant or nasal vowel precedes ne, it is likely that 2Neg will be
used. In addition, ne retention is favored when there is no preceding phonological
environment, such as in the case of imperatives.
GoldVarb 2001 did not find following phonological environment to be significant,
which is attributed to a lack of significant difference in ne retention rates within this factor
group. Table 16 gives the overall distribution of 2Neg and Type D of 1Neg according to
following phonological environment.
50
Table 16. 2Neg and 1Neg distribution according to following phonological environment. Environment 2Neg 1Neg Type D Total Consonant 117 (16.93%) 574 (83.07%) 691 (100%) Vowel 78 (14.97%) 443 (85.03%) 521 (100%) Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%)
A combination of the results for preceding and following phonological environment
suggests that the surrounding phonological environment might determine—at least to some
extent—whether ne is present or absent. These results are shown in Table 17. An example of
each phonological environment is also provided.
Table 17. Ne retention according to surrounding phonological environment. Environment 2Neg 1Neg Type D Total Consonant_Consonant (C_C) [elle (ne) veut pas] 'she does not want to'
18 (25.00%) 54 (75.00%) 72 (100%)
Consonant_Vowel (C_V) [il (n') est pas là] 'he is not there'
27 (36.49%) 47 (63.51%) 74 (100%)
Vowel_Consonant (V_C) [tu (ne) sais pas] 'you do not know'
72 (14.91%) 411 (85.09%) 483 (100%)
Vowel_Vowel (V_V) [tu (n') es pas là] 'you are not there'
42 (10.07%) 375 (89.93%) 417 (100%)
Nasal_Consonant (N_C) [on (ne) fait pas ça] 'one does not do that'
9 (20.45%) 35 (79.55%) 44 (100%)
Nasal_Vowel (N_V) [on (n') est pas là] 'one is not there'
5 (25.00%) 15 (75.00%) 20 (100%)
Ø_Consonant (Ø_C) [(ne) parlez pas] 'do not speak'
18 (19.57%) 74 (80.43%) 92 (100%)
Ø_Vowel (Ø_V) [(n') achetez pas ça] 'do not buy that'
4 (40%) 6 (60%) 100 (100%)
Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%) In order to determine whether selected phonological environments were influential on
ne retention rates, a series of chi-square tests was performed. Table 18 shows clearly that, in
many cases, the leading and following phonological environments interact and influence,
together, whether ne will be present or absent. The abbreviations shown in Table 17 (e.g.,
51
C_C, C_V, V_C, and so forth) have been used in the remainder of the discussion of
surrounding phonological environment.
Table 18. Interaction of selected phonological environments. Environments Chi-square results C_C vs. V_C χ2 (1, N = 555) = 4.69, p = .05 C_C vs. V_V χ2 (1, N = 489) = 12.71, p = .05 C_V vs. V_C χ2 (1, N = 557) = 20.45, p = .05 V_C vs. V_V χ2 (1, N = 900) = 4.73, p = .05
Let us consider, for example, C_C and V_C environments, both of which were
included in the "following consonant" category. Table 17 suggests that ne retention is
dependent upon whether a vowel or consonant precedes the ne position (25% for C_C and
14.91% for V_C) and the results given in Table 18 show the significance of the distribution.
A comparison of C_V and V_V environments—which had drastically different ne
retention rates (36.49% for C_V and 10.07% for V_V; see Table 17)—also reveals that the
presence or absence of pre-vocalic ne is determined by the sound (a consonant or vowel) that
precedes it. In addition, the data suggest that ne retention rates with preceding vowels are, at
least to some extent, dependent upon the phonological environment that follows the ne
position. The higher rate of ne retention for V_C environments (14.91%; see Table 17)
compared with V_V environments (10.07%) was also found to be significant (see Table 18).
What emerges from the analysis of selected surrounding phonological environments is
that the variable presence or absence of ne is determined to some extent by both the
environment immediately preceding the ne position and the environment that immediately
follows it. Specifically, ne retention rates are significantly lower in intervocalic position than
when ne follows a vowel and precedes a consonant.
4.3.3 Analysis of the Second Coding: Morphosyntactic and Lexical Environment
The second part of the analysis of the second coding considers Neg2 type (i.e., pas vs.
Neg2 other than pas), subject type (e.g., NP, pronoun, [- overt] subject), and sentence type
(e.g., declarative, interrogative, imperative). After reviewing the one-level binomial analysis
52
(see Table 19 below for factor weights and Appendix F for full results), it was obvious that
by separating the two phonological factor groups from the other structural factor groups
considered in the second coding, the results fit well within the expected model, according to
the criteria explained by Young and Bailey (1996) and Tagliamonte (2006).
Table 19. Factor weights from one-level binomial analysis for Neg2 type, subject type, and sentence type: second coding. Factor Group Weight Neg2 Type
pas .491 Neg2 other than pas .535
Subject Type NP .895 Pronoun .446 [- overt] subject .580
Sentence Type Declarative .505 Interrogative .433 Imperative .511
A step-up/step-down analysis was run for the three factor groups considered in this
part of the second coding (see Table 25 below for significant Goldvarb scores and Appendix
G for full results). After reviewing all runs reported in this part of the analysis, subject type
emerged as a significant factor. GoldVarb 2001 did not, however, find Neg2 type or sentence
type to be significant. Incidentally, when run independently of phonological environment,
factor weight constraints within the subject type factor group did not change (as they had in
the step-up/step-down procedure in section 4.3.1), which further supports the hypothesis that
subject type and preceding phonological environment are inextricably linked.
Neg2 Type
The data shown in Table 19 suggest that ne retention does not depend on the type of
Neg2 present. Table 20 shows clearly that ne retention rates are similar among most of the
different Neg2 types. Nonetheless, the slight difference in ne retention rates between pas and
Neg2s other than pas suggests that ne occurs at relatively higher frequencies with items that
occur less frequently in discourse (Coveney, 1996; Hansen & Malderez, 2004).
53
Table 20. Distribution of 1Neg Type D and 2Neg according to Neg2 type: second coding. Neg2 2Neg 1Neg Type D Total pas 143 (14.91%) 816 (85.09%) 959 (100%) Neg2 other than pas 52 (20.55%) 201 (79.45%) 253 (100%) Total 195 (16.09%) 1,017 (83.23%) 1,212 (100%) Although GoldVarb 2001 did not find Neg2 type to be significant, the distribution of
2Neg and Type D of 1Neg was found to be significant by a chi-square procedure when all
single Neg2s (e.g., pas, rien, jamais, etc.) were collapsed and tested against the negations in
which more than one Neg2 had been used: χ2 (1, N = 1,210) = 5.04, p = .05. This finding
strongly suggests that ne retention is favored when more than one Neg2 is used in a negation
(e.g., plus rien, plus personne, etc.). This distribution is shown in Table 21.
Table 21. Single vs. Multiple Neg2s: second coding. 2Neg 1Neg Type D Total Single Neg2s 186 (15.72%) 997 (84.28%) 1,183 (100%) Multiple Neg2s 9 (31.03%) 20 (68.97%) 29 (100%)
plus rien 2 7 9 plus personne 2 3 5 personne...rien 2 0 2 pas que 0 3 3 pas rien 0 3 3 jamais que 2 0 2 plus jamais 1 0 1 jamais rien 0 2 2 jamais personne 0 2 2
Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%)
Although retention rates are higher when more than one Neg2 is used, the results of
this analysis reveal that the variable use of ne does not normally depend on the type of
Neg2.7 This finding is in itself rather important since it suggests that ne non-retention has
become more or less generalized regardless of which second negative is present. Nonethele
it is important to note that a corpus with more Multiple Neg2 tokens would provide a more
solid basis for making this conclus
ss,
ion.
7 This finding does not corroborate results reported in Hansen & Malderez (2004), who found that more frequently occurring Neg2s (principally pas) were less likely to co-occur with ne than less frequently occurring Neg2s.
54
Sentence Type
The results of the step-up/step-down procedure for the second coding also reveal that
the variable use of ne does not depend on sentence type. The distribution does, however,
reveal, once again, to what extent ne non-retention has been generalized. Table 22 divides
tokens of negation by sentence type: declarative, interrogative, and imperative.
Table 22. Ne retention according to sentence type: second coding. Sentence type 2Neg 1Neg Type D Total Declarative 166 (16.27%) 854 (83.73%) 1,020 (100%) Interrogative 10 (10.75%) 83 (89.25%) 93 (100%) Imperative 19 (19.19%) 80 (80.81%) 99 (100%) Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%) The data shown in Table 22 indicate that ne retention rates are similar in all three
environments. It is therefore reasonable to assume that the presence or absence of ne does not
depend on the type of sentence; however, the analysis of a larger corpus with more tokens of
negative interrogative and imperative sentences might prove insightful.
Subject Type
Tokens of negation with [+ overt] subjects were also considered. Table 23 divides [+
overt] subject type into two main categories: noun phrases (NP) and pronouns. In addition to
substantives, pronominalized negative particles that function as subjects (e.g., personne, rien,
and so forth) were also included in the NP category since they are, historically, substantives
(Ewert, 1969; Ashby, 1981; Rickard, 1989). Although pronominalized negative particles
often function as objects (e.g., Je n'entends personne 'I hear no one'), as subjects, they appear
to be more closely related to nouns and, therefore, to their historical meaning (Ashby, 1981;
Grevisse, 1993, p. 1,076). The pronoun category includes all instances of clitic subject
pronouns (e.g., je, tu, on, etc.), as well as relative pronouns (e.g., qui and ce qui) and
demonstrative pronouns (e.g., ça and cela).
There exists a noticeable difference in ne retention rates between NPs and pronouns.
Table 23 shows clearly that there is a preference to retain ne when an NP is present (59.55%),
55
while ne retention does not appear to be preferred when a subject pronoun is used (12.01%).
A chi-square was calculated, which found the difference in ne retention rates between NPs
and pronouns to be significant: χ2 (1, N = 1,119) = 140.09, p = .05. These results corroborate
the findings of Ashby (1981), Coveney (1996), and Hansen & Malderez (2004).
Table 23. Distribution of 2Neg and 1Neg Type D according to [+ overt] subject type: second coding. [+ Overt] Subject 2Neg 1Neg Type D Total NP 53 (59.55%) 36 (40.45%) 89 (100%)
Noun8 45 (60.81%) 29 (39.19%) 74 (100%)
Negative particle 8 (53.33%) 7 (46.67%) 15 (100%) Pronoun 123 (12.01%) 901 (87.99%) 1,024 (100%)
je 60 (12.79%) 409 (87.21%) 469 (100%) tu 10 (6.71%) 139 (93.29%) 149 (100%) il 8 (16.00%) 42 (84.00%) 50 (100%) elle 4 (14.29%) 24 (85.71%) 28 (100%) on 9 (15.79%) 48 (84.21%) 57 (100%) ce 4 (2.76%) 141 (97.24%) 145 (100%) nous 2 (100%) 0 (0.00%) 2 (100%) vous 2 (11.11%) 16 (88.89%) 18 (100%) ils 1 (10.00%) 9 (90.00%) 10 (100%) elles 1 (33.33%) 2 (66.67%) 3 (100%) ça9 9 (14.06%) 55 (85.94%) 64 (100%) qui10 11 (40.74%) 16 (59.26%) 27 (100%)
[+ Overt] Subject Total
176 (15.81%) 937 (84.19%) 1,113 (100%)
Fifteen examples of Neg2s were included in the NP subject category, eight of which
(53.33%) co-occurred with ne. A closer analysis of the examples of Neg2 subjects revealed
that [personne(subject) + Neg2] was used in four of the eight tokens of 2Neg in this category; in
fact, not one example of [personne(subject) + Neg2] was counted as 1Neg. The remaining four
tokens of 2Neg were examples of personne and rien (two examples each). The data suggest
that when more than one Neg2 is used in negation, ne retention is favored, provided that one
of the Neg2s present is the subject of the clause.
8 One instance of les miennes (2Neg) was included as a NP. 9 One instance of cela (2Neg) was included with tokens of ça. 10 Four instances of ce qui (three 2Neg and one 1Neg) have also been included in this category.
56
Ne retention according to the various types of pronouns identified in the corpus was
also considered. Although ne retention is not favored when a pronoun is present in general
(see Table 23), ne retention rates vary depending upon which pronoun is used. Table 24
divides subject pronouns into clitic and non-clitic.11
Table 24. Distribution of 2Neg and 1Neg Type D with clitic and non-clitic subject pronouns: second coding Pronoun type 2Neg 1Neg Type D Total Clitic 103 (11.04%) 830 (88.96%) 933 (100%) Non-clitic 20 (21.98%) 71 (77.02%) 91 (100%) Total 123 (12.01%) 901 (87.99%) 1,024 (100%)
The data shown in Table 24 indicate that ne co-occurs more frequently with non-clitic
subject pronouns than with clitic subject pronouns, which corroborates results reported in
previous studies (Coveney, 1996; Armstrong, 2002; Armstrong & Smith, 2002; Hansen &
Malderez; 2004). A chi-square test revealed that the distribution of 2Neg and 1Neg Type D in
Table 24 was significant: χ2 (1, N = 1,024) = 9.39, p = .05.
One possible explanation for this difference is that relative pronouns—which
account for 27 of the 91 tokens of non-clitic pronouns—occur less frequently than other types
of pronouns. If one accepts the hypothesis that frequency of occurrence determines, at least to
some extent, the likelihood that ne will be absent or present (see Coveney, 1996; Hansen &
Malderez, 2004), it seems reasonable to assume that the low frequency of qui and ce qui in
this corpus contributes to the high rate of co-occurrence with ne.
In addition, there is a noticeable difference in ne retention rates between qui and ça
(40.74% for qui and 14.06% for ça). This may be explained by the functional difference
between relative and demonstrative pronouns. The demonstrative pronoun ça is often used in
pre-formed sequences, in both affirmative and negative phrases (e.g., ça va or ça [ne] va
pas), which may account for the low rate of ne retention. Relative pronouns, however, are
usually present in novel sentence structures and refer to a NP or an idea introduced in the
11 Non-clitic pronouns include instances of demonstrative and relative pronouns.
57
preceding clause, which appears to favor ne retention (see Coveney, 1996).
In light of Armstrong & Smith's (2002) suggestion that [- overt] subject environments
focus the speaker's attention on the negation and result in higher ne retention rates,
environments in which no overt subject is present were explored. In this analysis, only
imperatives were considered (see 4.3.1). Ne retention rates are relatively high in the [- overt]
subject category (19.19%).
In [+ overt] subject environments, ne retention appears to be disfavored (see Table
23), yet it is favored—at least to some extent—in [- overt] subject environments (see Table
24). A chi-square was calculated for these two environments and found that this factor was
indeed significant: χ2(1, N = 1,240) = 7.19, p = .05. In addition, GoldVarb 2001 found that
the factor group "subject type" (i.e., NP, pronoun, [- overt] subject) to be significant. These
results are shown in Table 25.
Table 25. Distribution of 2Neg and 1Neg Type D according to subject type: second coding. Subject environment
2Neg 1Neg Type D Total GoldVarb Score
NP 53 (59.55%) 36 (40.45%) 89 (100%) .899 Pronoun 123 (12.01%) 901 (87.99%) 1,024 (100%) .455 [- overt] subject 19 (19.19%) 80 (80.81%) 99 (100%) .583 Total 195 (16.09%) 1,017 (83.91%) 1,212 (100%) — The data shown in Table 25 reveal that ne retention is strongly favored when the
subject of the verb is an NP. Incidentally, a GoldVarb 2001 score of .899 is extremely high,
which indicates that the rule (i.e., the use of 2Neg) will be applied very often in this particular
environment; yet, since the GoldVarb 2001 score for pronoun is less than .500, ne retention is
disfavored in this environment. In addition, the GoldVarb 2001 score for [- overt] subject
environment indicates that the rule will be applied since the score is over .500. Although
these results are not surprising when compared to those reported by Ashby (1981), Coveney
(1996), and Hansen & Malderez (2004), they are indicative of the relationship between the
discourse of IRC and everyday conversational speech. This in itself suggests that participants
58
are writing in this communicative environment as they would speak in informal settings, at
least as far as the variable ne is concerned. However, writing as it is traditionally understood
(e.g., letters, literature, print publications, and so forth) must be distinguished from the
spontaneous communication that occurs in this synchronous text-based environment.
Il faut sans doute se méfier de la vue étroite et idéalisée que l'on a en général
de l'écrit, identifié au texte publié, élaboré dans la durée et corrigé par des
professionnels. Il est sans doute difficile de trancher entre une influence de
l'oral sur l'écrit et un rapprochement déterminé par l'élargissement du domaine
d'usage de l'écrit. De plus, le partage entre la contamination spontanée et le
recours intentionnel à des effets d'oralité est pratiquement impossible à
effectuer. (Anis, 1999, p. 75)12
4.3.4 Interaction of Subjects and Phonological Environment
Although the analysis presented in 4.2 highlights that subject type and phonological
environment interact to such an extent that it is necessary to consider the two separately when
performing a multivariate analysis such as that performed by GoldVarb 2001, upon review of
the results of the analysis it became clear that this interaction could not simply be left
unexplored. A cross-tabulation of subject type and preceding phonological environment
revealed that no less than 874 pronouns, which appear to disfavor ne retention (see Table 25),
were counted in the preceding vowel category. Table 23 provides the distribution of 2Neg
and Type D of 1Neg with pronouns ending in a vowel (or vowel sound) according to
following phonological environment (e.g., consonant or vowel).
12 Translation: "We must be weary of the strict and idealized view of writing, which is associated with published texts that are elaborated over time and corrected by professionals. It is without a doubt difficult to distinguish between an influence of spoken language on written language and a rapprochement that is determined by the expansion of the use of writing. In addition, it is practically impossible to make a distinction between spontaneous contamination and the intentional use of oral characteristics."
59
Table 26. Pronoun with final vowel and following phonological environment. Following environment
2Neg 1Neg Type D Total
Consonant 67 (14.19%) 405 (85.81%) 472 (100%) Vowel 29 (7.21%) 373 (92.79%) 402 (100%) Total 96 (10.98%) 778 (89.02%) 874 (100%) It is clear that the phonological environment following a pronoun ending in a vowel is
influencing ne retention. Table 23 shows that ne occurs at a higher rate when a consonant
follows than when a vowel follows. A chi-square procedure found the distribution to be
significant: χ2 (1, N = 874) = 10.82, p = .05. One possible explanation for this difference is
the fact that contractions are very often made between certain pronouns and a following
vowel (or vowel sound). Indeed, contractions are required with je and ce when a vowel
follows (e.g., j'ai or c'est), while other more informal (optional) contractions are made, for
example, with tu (e.g., t'es). Since such contractions occur rather frequently in affirmative
phrases, and it can be argued that speakers most usually operate with the [pronoun + verb]
sequence (Ashby, 1981; Coveney, 1996), it is not surprising that in negative phrases, ne
retention rates are extremely low in the intervocalic position (approximately one-half the
overall retention rate).
Ne retention with pronouns ending in a consonant (e.g., il, elle, ils, elles) was also
considered. Table 27 gives the distribution of 1Neg Type D and 2Neg for these pronouns
according to following phonological environment.
Table 27. Pronoun with final consonant and following phonological environment. Following environment
2Neg 1Neg Type D Total
Consonant 4 (9.09%) 40 (90.91%) 44 (100%) Vowel 11 (23.40%) 36 (76.60%) 47 (100%) Total 15 (16.48%) 76 (73.52%) 91 (100%)
Although a chi-square test did not find the distribution shown in Table 27 to be
significant, the distribution suggests that ne retention is not common in the context of a
pronoun with a final consonant followed by another consonant; yet, the results shown in
60
Tables 17 and 18 (section 4.3.2) indicate that ne retention is favored in C_C environments.
This discrepancy is most likely explained by the fact that examples of C_C environments
were predominately tokens of NPs, which strongly favor ne retention.
Ne retention also appears to occur more often when a vowel follows a pronoun ending
in a consonant. This may suggest that ne is used in order to avoid liaison is certain cases. In
addition, an analysis of environments in which liaison would result in a [z] (e.g., vous
[n']avez pas, ils [n']ont pas) was preformed and revealed that ne was retained in 33.33% of
such instances.
One final observation was made regarding an environment that has traditionally been
omitted from studies of ne: the clitic on followed by a vowel (or vowel sound). This
environment has been considered difficult (if not impossible) to study in spoken French since
it is unclear whether [n] is the result of liaison or pre-vocalic ne (e.g., on n'a pas de pommes
[õnɑpɑdəpɔm] 'we don't have any apples'). This obstacle does not exist in the chat
environment since the written transcription of the discussion is available for analysis.
Fifty-seven tokens of on were found in the corpus, fourteen of which were followed
by a vowel sound. Of these fourteen tokens, only one (7.14%) co-occurred with ne. Although
the limited number of tokens does not allow for any definite conclusions to be drawn from
this analysis, the results do provide some evidence that the [n] observed in spoken discourse
might be the result of liaison rather than pre-vocalic n'. It is also possible that speakers
themselves do not know whether the [n] is pre-vocalic n' or the result of liaison. If this is true,
it appears that they have generalized non-retention to this environment in synchronous
French-language chat.
4.3.5 Summary of Results: Second Coding
The results from the two separate analyses of phonological environment and Neg2
type, subject type, and sentence type suggest that both the phonological environment
61
immediately preceding the ne position and subject type are significant factors. The data
reported in 4.3.2 indicate the ne retention is disfavored when a vowel preceded the ne
position, yet it is favored when a consonant or a nasal vowel is present. In addition, it appears
that 2Neg is preferred when there is no preceding phonological environment. The results
shown in 4.3.3 suggest that the absence of ne is rather generalized and does not depend upon
the type of Neg2 used or sentence type. The one factor that emerges as significant in this part
of the analysis is subject type; specifically, ne retention is very strongly favored when an NP
is present, while 1Neg Type D appears to be preferred with subject pronouns. The retention
of ne also seems to be favored in [- overt] subject environment, which corroborates the
findings reported in 4.3.2 concerning the factor "no preceding phonological environment."
Last, a review of the data revealed that the vast majority of tokens counted as
preceding vowels in 4.3.2 were instances of pronouns (see 4.3.4), which disfavor the use of
2Neg. It was found that pronouns ending in consonants were more likely to co-occur with ne
than those that end in vowels. Moreover, a large percentage of tokens that had been counted
in the preceding consonant category were found to be NPs, which strongly favor ne retention.
Although these results are far from conclusive, type of subject emerges as the most influential
internal linguistic factor, while phonology is most likely a contributing or underlying factor in
the variation.
4.4 Discursive-Pragmatic Effect
4.4.1 Overview of Style-Shifting
The results shown in 4.3 indicate that a number of structural and phonological factors
determine—at least to some extent—whether ne is present or absent; yet, it is possible that
other, non-structural, factors influence the variation as well. Ashby (1981) posited that
discourse topic (i.e., subject of discussion) was an influential factor in ne retention. He found
that ne retention co-occurred frequently with serious topics, such as the punishment of
62
children and religion, as well as with the telling of general truths (e.g., proverbs, common
sayings, etc.). In addition, Armstrong & Smith (2002) advanced the hypothesis that ne could
be used in order to reinforce or add emphasis to a negation, and Coveney (1996, pp. 88-89)
reported an extreme style shift by one 35-year-old male informant, presumably because the
informant felt at one point that he should speak more clearly so that non-native speakers of
French could understand him more easily. However, discussions and analyses of the
intraspeaker dimension and discursive-pragmatic effect remain anecdotal in previous studies.
Indeed, Armstrong (2002) is one of the few researchers to have undertaken qualitative
analyses of the variable ne along the intraspeaker, stylistic dimension.
Armstrong's (2002) quantitative results do not indicate an observable pattern or
system since some of the speakers retained ne at higher rates in interview style, yet others
used ne at higher rates in conversation style. What emerges from Armstrong's analysis is that
style-shifting was occurring on the micro-level; in other words, individual speakers appeared
to produce tokens of ne during brief episodes of formal-style discourse in both conversation
and interview styles, which suggests a diminished importance of the traditional binary
paradigm of formal vs. informal.
Several of the stylistic effects produced by the Dieuze informants in
conversation style through their use of ne . . . contradict a simplex formal-
informal analysis of style variation. Micro-style variation . . . is reflected in the
use of ne through a reduction in the degrees of style shift, since the 'formal'
episodes in conversation style call for the use of ne quite frequently relative to
interview style. (Armstrong, 2002, p. 171)
In light of Bell's (1984, 2001) theory of variation as audience design and Armstrong's
(2002) analysis of negative particle use along the intraspeaker dimension, I have chosen to
explore the ne retention rates of six different chat participants in order to compare the
63
variation found in individuals with that of the corpus as a whole. Bell (2001) defines audience
design as "a strategy by which speakers draw on a range of linguistic resources available in
their speech community to respond to different kinds of audiences" (p. 145) and argues that
style shift is a responsive, yet active, behavior (pp. 143-144). Figure 3 provides an illustration
of how variation as audience design functions.
Figure 3. Bell's model of audience design (adapted from Bell, 2001, p. 142).
(1) Group has its own identity, evaluated by self
and others
(2) Group differentiates its language from others': "social," or inter-speaker
variation
(3) Group's language is evaluated by self and
others: linguistic evaluation
(4) Others shift relative to group's language: "style," or intra-speaker variation
(markers)
Bell (1984, 2001) assumes that each group (i.e., speech community) is seen as an
independent entity (1) that has established a set of linguistic norms that are particular to that
group (2). The language used within the group (i.e., among group members) is constantly
evaluated (3) by both the group as a whole and individual members. Those who participate in
the group shift their use of language relative to the expected norms of the group (4). Thus, the
underlying principal of variation as audience design presupposes that "style is oriented to
people rather than to mechanisms or functions" (Bell, 2001, p. 141).
Of primary interest to the analysis of discursive-pragmatic effect in the present study
is the notion that style-shifting can occur according to the topic of discussion, a shift that
64
"derives its meaning and direction of shift from the underlying association of topics or setting
with typical audience members" (Bell, 2001, p. 146). In addition, style-shifts may occur when
"the individual speaker . . . uses language resources often from beyond the immediate speech
community" (Bell, 2001, p. 147). As the analysis presented earlier has indicated, there exists
an overwhelming preference for 1Neg Type D, which suggests that the discourse of
synchronous CMC is closely associated with that of everyday conversational speech.
Therefore, those who engage in chat discussions shift their style of discourse in the direction
of everyday conversational speech, indicated here by the overwhelming preference for single-
particle negation. Yet instances of 2Neg were observed in this analysis, and although patterns
were found concerning a number of syntactic factors, an analysis of tokens of 2Neg might
reveal other underlying factors in the variation; specifically shifts in the topic or tone of the
discussion that led to the use of 2Neg.
4.4.2 Overview of Selected Participants
The six participants chosen were the top six contributors in the corpus (i.e., they
produced the highest number of turns). Additionally, each of the six participants was present
on at least three occasions during data collection. These six participants alone produced
11,438 words (over 14% of the data) and approximately 16.21% of all tokens of negation
(both 1Neg Type D and 2Neg). Table 28 shows clearly that the overall rate of 2Neg for these
six participants (16.92%) is approximately equivalent to the overall frequency of ne retention
(16.77%) in this corpus. It therefore seems reasonable to assume that ne use among these six
participants is representative of the corpus.
65
Table 28. Distribution of 2Neg and 1Neg Type D for top six contributors. Participant 2Neg 1Neg D Total Neg <angelina> 1 (1.67%) 59 (98.33%) 60 (100%) <salizar> 12 (44.44%) 15 (55.56%) 27 (100%) <prue> 2 (7.41%) 25 (92.59%) 27 (100%) <ange> 3 (9.38%) 29 (90.63%) 32 (100%) <bruluin> 1 (4.76%) 20 (95.24%) 21 (100%) <nomade> 15 (44.12%) 19 (55.88%) 34 (100%) Total 34 (16.92%) 167 (83.08%) 201 (100%)
Along the intraspeaker dimension, rates of ne retention are unevenly distributed.
While no single participant categorically omitted ne, <angelina> and <bruluin> produce only
one token of 2Neg. These findings are not, however, uncommon (see Armstrong, 2002). In
addition, two participants produce an extremely high number of 2Neg tokens (<salizar> and
<nomade>). How can the differences in ne retention rates shown in Table 26 be explained?
What factors or influences can we identify? In which contexts is ne being used? In order to
address these questions, it is necessary to examine the tokens of negation produced by these
six participants.
4.4.3 Definitions and Examples of Discourse Styles
A preliminary analysis of the sub-sample of data considered in this section revealed
that tokens of 2Neg often correlated with a certain number of discourse topics, including
arguments, jokes, imitations/role-plays, general truths, and policing. In order to test whether
the topic of discourse is in fact influential, a statistical analysis was performed using
GoldVarb 2001. The tokens of negation (N = 201) produced by <angelina>, <salizar>,
<prue>, <ange>, <bruluin>, and <nomade> were first re-coded according to the internal
linguistic factors discussed in 4.2, including subject type (NP, pronoun, [- overt] subject),
preceding and following phonological environment, and sentence type (declarative,
interrogative, imperative). In addition, the tokens were coded according to four general
categories of "discourse style": ludic, emphatic, proverbial, and explanatory.
66
Ludic discourse style includes examples of jokes, imitations, and role-playing, as
illustrated by Excerpts 11 and 12. Although it was not always immediately apparent that a
message had been intended to convey humor, it was often the reaction of other participants
that made classification possible. For example, smileys and other graphemes are often
included in messages in order to convey emotion (Pierozak, 2003c & 2006). (In the following
excerpts of data, sic has not been used to indicate that a grammatical, orthographic, or other
type of mistake has been merely reproduced. Tokens of negations have been underlined and
English translations in italics have been provided below data excerpts.
Excerpt 11. <ange_away> Eliot de koi tu te plains krib a dit ke ct pa toi le boulet dont il parlait [...] <ange_away> :p <ange_away> Eliot what are you complaining about krib said that you weren't the idiot he was talking about [...] <ange_away> :p Excerpt 12. <Anonyme952046> yen a pa ki chatte un pe la? <Salizar> Je ne pense pas. <Salizar> (Donc je ne suis pas) <Anonyme952046> aren't there any who are chatting a little? <Salizar> I do not think [so]. <Salizar> (Therefore I am not)
The second line of Excerpt 11 includes the smiley :p, which represents a stuck-out-
tongue smiley face, which is most usually associated with humor, much as sticking one's
tongue out while smiling (jokingly) would be in face-to-face communication. Although
Excerpt 12 does not include any examples of such graphemes, it is nonetheless a clear-cut
example of ludic discourse. <Salizar> responds to <Anonyme952046> by quoting Descartes'
statement "je pense, donc je suis" (from his Discours de la méthode, 1637), but turns the
phrase in adding the negation.
Emphatic discourse style normally occurs during arguments and disputes. This style is
also used during "policing" (i.e., when a participant is warned or punished for his or her
67
behavior, which is usually related to the use of profanity or unwelcome statements). Excerpt
13 provides another example of Emphatic discourse style since <angelina> does not appear to
appreciate <zidzid>'s comment about HELENE33|repo's breathing troubles.
Excerpt 13. <angelina> est alors ta respiration ca va ma HELENE33|repo ? <zidzid> elle etouf <angelina> zidzid chute un peu de respect [...] <angelina> parle mieu ou tu gercle te moque pas des gents <angelina> so how's your breathing my HELENE33|repo ? <zidzid> she's suffocating <angelina> zidzid shhhh a little respect [...] <angelina> speak better or you'll get it don't make fun of people Proverbial style is similar to Laberge & Sankoff's (1977) "morals and truisms"
category. Examples of general truths, rules, and proverbs have been included in this category.
Although a certain number of the examples counted as proverbial have a somewhat ludic
function, I believe that it is important to distinguish between ludic style, which involves
jokes, and proverbial style, which exposes truths and generalities. Excerpts 14 and 15 provide
examples of proverbial discourse style.
Excerpt 14. <Salizar> Comme je le dis toujours : Un couple qui ne se dispute plus c'est un couple qui s'en fout <Salizar> As I always say: A couple that no longer fights, is a couple that does not care Excerpt 15. <nomade> il n y a aucune obligation a distibuer des pelles et des kiss <nomade> there is no obligation to give out kisses Although Excerpt 14 is much more closely related to a proverb than 15, both
examples have the same basic function: they provide general information and value
judgments of a situation. Such examples were categorized as proverbial style.
Explanatory discourse style is what one might be tempted to label "normal
conversation style." General explanations, observations, descriptions, and questions have
been included in the Explanatory discourse style category. Although this may appear to be a
68
vast (even vague) definition, a preliminary analysis of the data did not find that this type of
discourse was any more common than ludic, emphatic, or proverbial style. Excerpts 16, 17,
and 18, provide examples of explanatory discourse style.
Excerpt 16. <Prue> n'empêche que c'est pas forcément vrai, mais pas faux non plus Hugo :) <Prue> albeit it's not necessarily true, but not false either Hugo :) Excerpt 17. <ange_away> je l'ai trouvée <ange_away> la robe <ange_away> :p <ange_away> ah nan <ange_away> g ^pa trouvé <ange_away> i found it <ange_away> the dress <ange_away> :p <ange_away> ah no <ange_away> I didn't find it Excerpt 18. <angelina> er je suis desoler je pouvais rien faire rien ne marcher pardon les gents <angelina> er i'm sorry i couldn't do anything nothing was working sorry guys The three examples above demonstrate clearly that explanatory discourse style is a
rather general, indeed explanatory, type of discourse. It differs from the other three discourse
styles identified in the corpus in that only linguistic information is communicated. In other
words, no extra- or meta-linguistic information—such as humor, anger, or judgment—is
conveyed in the message.
4.4.3 Results
The results of the GoldVarb 2001 analysis for the 201 tokens of negation considered
in this part of the analysis indicate that, once again, subject type emerges as a significant
factor group.
69
Table 29. Distribution of 2Neg and 1Neg Type D according to subject type. Subject type 2Neg 1Neg Type D Total GoldVarb
score NP 10 (76.92%) 3 (23.08%) 13 (100%) .962 Pronoun 20 (12.12%) 145 (87.88%) 165 (100%) .469 [- overt] 4 (17.39%) 19 (82.61%) 23 (100%) .282 Total 34 (16.92%) 167 (83.08%) 201 (100%) — It is clear that ne retention is favored with NP subjects, while it is disfavored with
subject pronouns. However, in this sub-sample of data, [- overt] subjects appear to disfavor
ne retention. Phonological environment (both preceding and following) and sentence type
were not found to be statistically significant, which corroborates the results reported earlier in
4.2.3 and 4.2.4. Low cell counts for second negatives other than pas made a quantification of
this factor impossible. Overall, the results from the six participants selected for analysis
appear to reflect what was found for the whole corpus, and it is therefore reasonable to
assume that the results concerning discourse style will be rather insightful.
The four different types of discourse style co-occurred with ne at various frequencies,
yet ne was neither retained nor omitted categorically in any one of them. GoldVarb 2001
found discourse style to be statistically significant. Table 31 provides Varbrul scores for the
independent variables in this factor group and ne retention rates according to discourse style
for the six participants chosen for analysis.
Table 30. 2Neg and 1Neg Type D distribution according to discourse style. Discourse style 2Neg 1Neg Type D Total Varbrul Score Explanatory 2 (1.79%) 110 (98.21%) 112 (100%) .130 Proverbial 4 (50.00%) 4 (50.00%) 8 (100%) .865 Emphatic 5 (22.73%) 17 (77.27%) 22 (100%) .886 Ludic 23 (38.98%) 36 (61.02%) 59 (100%) .930 Total 34 (16.92%) 167 (83.08%) 201 (100%) —
It is clear that discourse style determines—at least in part—whether ne is retained. Ne
retention is strongly disfavored in the explanatory discourse style, while it is favored in
proverbial, emphatic, and ludic styles. In addition, 23 of the 34 examples of 2Neg (67.65%)
were counted in the ludic category. This should not be surprising, however, given the
70
"increased opportunities for language play" (Herring, 1999) provided by the synchronous
chat environment (see also Pierozak, 2003a). The playful aspect of this type of
communication may even be a reason for its popularity. According to Herring (1999):
On-line surveys reveal that humorous messages are the most highly
appreciated types of messages in computer-mediated environments, even in
those one would not necessarily characterize as 'recreational'. . . . The
potential for humorous play inherent in the . . . computer-mediated
environment constitutes one of the biggest attractions of CMC for many users.
Table 31 also shows that emphatic discourse style favors ne retention, which supports
the claim that ne can be used to add secondary emphasis to a negation (Armstrong & Smith,
2002), especially during disputes and arguments. Further, it should not be surprising that ne
retention is favored in proverbial style since it is reasonable to assume that a speaker might
be more careful in his or her speech when citing a proverb or telling a general truth. This
provides further evidence for the hypothesis that more "standard" forms and structures are
used when more attention is paid to speech (Labov, 1972).
In addition, following Bell's (2001) model of audience design, it can be argued that
"the individual speaker . . . uses language resources often from beyond the immediate speech
community" (p. 147). Although ne retention is generally disfavored in chat environments, the
analysis presented in the preceding paragraphs strongly suggest that group members (i.e.,
chat participants) draw upon their linguistic resources (i.e., 2Neg) "from beyond the
immediate speech community" in order to express humor and anger, and to imitate the
discourse of proverbs and truisms.
71
CHAPTER 5
CONCLUSION
5.1 Summary of Results
Comparisons of the discourse of synchronous French-language CMC and what one
would expect to observe in everyday conversational speech reveal a number of linguistic,
stylistic, and pragmatic similarities. In the case of ne use, synchronous CMC participants
omit ne from verbal negation at frequencies similar to those reported in previous research
concerning everyday conversational French (e.g., Coveney, 1996; Hansen & Malderez,
2004). In other words, ne is retained at variable rates depending on the linguistic and/or
stylistic environment (see ch. 4).
The results of this research suggest that subject type and phonological environment
are inextricably linked and that it may be necessary to consider these two factors separately
when doing a multivariate analysis of the use of ne. In this study, subject type and preceding
phonological environment emerged as influential factors in the variation of ne use when
separated. Although following phonological environment was not found to be significant, a
closer analysis of the phonological environment surrounding the ne position revealed that the
presence of ne does indeed depend on—at least to some extent—the sound following its
position. In addition, neither second-negative type nor sentence type was found to be
significant, which suggests that the variable use of ne is not influenced by these factors.
Overall, it appears that the type of subject present (e.g., NP or pronoun) or the lack of an
overt subject (e.g., imperatives) determines whether ne will be present or absent, while the
phonological environment preceding the ne position is an underlying and contributing factor,
72
specifically when a subject pronoun is used (see 4.3.4).
A qualitative analysis of tokens of negation produced by the six top contributors in the
corpus revealed that discourse style was a determining factor in the variation. While ludic,
emphatic, and proverbial styles appear to favor strongly the retention of ne, the presence of
ne is disfavored in explanatory style. These findings support Ashby's (1981) hypothesis that
discussion topic and discourse-pragmatic effect are influential factors as well as Armstrong's
(2002) suggestion that style-shifting occurs on a micro-level.
Results concerning selected syntactic factors reported in this study are not surprising;
overall, they corroborate results reported in previous research of ne in everyday
conversational speech (Ashby, 1981; Coveney, 1996; Hansen & Malderez, 2004). The
similarities between the discourse of synchronous French-language CMC and informal
spoken French are indicative of the relationship between these two types of discourse. In
other words, the language used in synchronous CMC resembles in many ways that of
informal spoken French; in particular the overwhelming preference for the omission of ne. It
therefore seems reasonable to say that synchronous CMC provides an environment in which
informal and non-traditional1 forms and structures are accepted as the norm.
Although some research (Ashby, 1981, 2001) has suggested that ne is disappearing
from the French language, another point of view maintains that we are far from seeing the
completion of such a change (see Hansen & Malderez, 2004). The present study supports the
latter of these two hypotheses since synchronous CMC participants use the variable ne in a
probabilistic way in certain syntactic and pragmatic environments. Indeed, the evidence
supporting the hypothesis that discourse style is a determining factor in the variation suggests
that ne has become an important pragmatic resource for synchronous CMC participants. How
is such a development possible? More specifically, which factors and influences can be
1 The term "non-standard" has been avoided since it could be argued that the absence of ne, although grammatically incorrect according to prescribed grammar, is, in fact, the standard in this type of communication.
73
identified to explain the evolution of ne from the default marker of negation to a
sociopragmatic tool capable of communicating humor and anger, among other extra-
linguistic information?
The literature on the history of the French language—specifically the development of
verbal negation—suggests that the phonetic weakening of Latin non rendered ne incapable of
bearing an energetic negation, which led to the grammaticalization of second-negatives and
the development of two-particle negation. Ashby (1981) highlights the effect of the
grammaticalization of second-negatives, which has contributed to the disposability of ne
since it is a redundant marker of negation. Further, "[i]t appears that this change . . . is given
impetus by another continuing change in French: the fusion of the clitic pronoun and the
verb" (Ashby, 1981, p. 686; see also Coveney, 1996; Fonseca-Greber & Waugh, 2003).
This final conclusion made by Ashby (1981) appears to be true for synchronous
French-language CMC. Recent research (Pierozak, 2003c; Williams, 2006; van Compernolle
& Williams, in press) has identified a number of fused [subject clitic + verb] sequences;
specifically c'est, tu es, and j'ai ('it is', 'you are', and 'I have', respectively). Electronic
environments (both synchronous and asynchronous) allow the user to abbreviate these
sequences; c'est [sɛ] becomes c [sɛ] (pronounced as the letter "c"), tu es (often contracted to
t'es [tɛ] in spoken French) becomes t [tɛ] (pronounced as the letter "t"), and j'ai [ʒɛ] becomes
g (pronounced as the letter "g").
Abbreviations such as these, which are often referred to as syllabograms,2 occur at
such high frequencies in affirmative sentences—presumably in order to save time when
engaging in synchronous CMC (see Pierozak, 2003b; van Compernolle & Williams, in
press)—that synchronous CMC users appear to operate principally with the [clitic + verb]
syllabogram in negative sentences as well. By reducing the [clitic + verb] sequence to a
2 Pierozak (2003c) proposed the French term syllabogramme.
74
single-syllable grapheme, the presence of ne is made impossible (see van Compernolle &
Williams, in press). If it is true that syllabograms are used in order to save time even in
negative sentences, than it is reasonable to assume that the use of more traditional forms (e.g.,
tu n'es pas) is deliberate. If such use is indeed deliberate, synchronous CMC users must be
aware of the pragmatic functions of ne—at least in this type of communication—and the
various discourse styles (e.g., ludic, emphatic, proverbial, and explanatory) that favor or
disfavor its presence.
5.2 Directions for Future Research
Although this study has undertaken a formal investigation of ne use in only one form
of synchronous CMC, informal observations suggest that ne retention rates are similar in
other synchronous CMC environments. In addition, recent research (van Compernolle &
Williams, in press) has suggested that ne use in IRC is very different from what can be
observed in discussion forums and moderated chat. These differences appear to be due—at
least in part—to the synchronicity of communication and the perceived level of formality of
the communication context. Future studies of the use of ne in CMC are needed in order to
determine to what extent the discourses of different types of synchronous and asynchronous
CMC are similar. In addition, more research on the effect of abbreviated orthographic forms
on the use of ne—among other linguistic variables—might prove insightful in the study of
sociolinguistic variation in CMC. Last, discourse-pragmatic effect on ne retention must be
explored further in both CMC and spoken French in order to understand more completely the
complex of linguistic and pragmatic factors that co-occur and co-vary with this variable.
75
76
APPENDIX A
GOLDVARB 200 CODING KEY (ALL DATA)
Factor Group 1: Neg2 Type 1 = pas 2 = Neg2 other than pas
Factor Group 2: Subject Type
q = NP e = Pronoun k = [- overt] subject
Factor Group 3: Preceding Phonological Environment
f = Vowel d = Consonant s = Nasal Vowel N = No preceding phonological environment
Factor Group 4: Following Phonological Environment
V = Vowel C = Consonant
Factor Group 7: Sentence Type
z = Declarative x = Interrogative c = Imperative
77
78
APPENDIX B
ONE-STEP BINOMIAL: FIRST CODING
Binomial Varbrul, 1 step ======================== Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. - One-level analysis only:One-level binomial analysis: Run # 1, 40 cells: No Convergence at Iteration 20 Input 0.143 Group Factor Weight App/Total Input&Weight 1: 2 0.530 0.21 0.16 1 0.492 0.15 0.14 2: e 0.454 0.12 0.12 q 0.892 0.60 0.58 k 0.511 0.19 0.15 3: f 0.447 0.13 0.12 s 0.544 0.22 0.17 d 0.486 0.31 0.14 N 0.863 0.22 0.51 4: C 0.524 0.17 0.16 V 0.468 0.15 0.13 5: z 0.543 0.16 0.17 c 0.172 0.19 0.03 x 0.446 0.11 0.12 Cell Total App'ns Expected Error 2qsVz 3 2 1.982 0.000 2qsCz 1 1 0.708 0.412 2qfVz 6 6 3.411 4.553 2qfCz 5 3 3.110 0.010 2qdVz 7 5 4.245 0.341 2qdVx 1 0 0.510 1.042 2qdCz 11 5 7.237 2.021 2qdCx 1 1 0.565 0.769 2kNCc 5 3 1.103 4.184 2esVz 4 0 0.657 0.786 2esCz 10 1 1.970 0.595 2efVz 64 7 7.519 0.041 2efVx 1 0 0.083 0.090 2efCz 104 12 14.819 0.625 2efCx 7 1 0.707 0.135
79
2edVz 10 3 1.347 2.346 2edVx 1 0 0.095 0.105 2edCz 11 2 1.789 0.030 2edCx 1 0 0.116 0.131 1qsVz 3 2 1.876 0.022 1qfVz 8 6 4.245 1.546 1qfCz 6 2 3.511 1.568 1qdVz 18 10 10.246 0.014 1qdVx 1 1 0.472 1.119 1qdCz 16 8 9.960 1.022 1qNVz 1 1 0.898 0.114 1kNVc 9 3 1.466 1.919 1kNCc 85 13 16.607 0.974 1esVz 10 1 1.443 0.159 1esCz 27 7 4.694 1.371 1esCx 6 0 0.748 0.854 1efVz 306 21 31.357 3.811 1efVx 32 2 2.294 0.041 1efCz 326 52 40.662 3.612 1efCx 35 2 3.077 0.414 1edVz 34 7 4.003 2.542 1edVx 2 1 0.166 4.583 1edCz 29 2 4.141 1.291 1edCx 3 0 0.304 0.338 1eNCx 2 2 0.857 2.666 Total Chi-square = 48.1946 Chi-square/cell = 1.2049 Log likelihood = -477.718
80
81
APPENDIX C
STEP-UP/STEP-DOWN ANALYSIS: FIRST CODING
Binomial Varbrul ================ Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. Threshold, step-up/down: 0.050001 # Stepping up: # Stepping up: ---------- Level # 0 ---------- Run # 1, 1 cells: Convergence at Iteration 2 Input 0.161 Log likelihood = -534.667 ---------- Level # 1 ---------- Run # 2, 2 cells: Convergence at Iteration 4 Input 0.160 Group # 1 -- 2: 0.576, 1: 0.480 Log likelihood = -532.419 Significance = 0.037 Run # 3, 3 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.445, q: 0.899, k: 0.583 Log likelihood = -483.652 Significance = 0.000 Run # 4, 4 cells: Convergence at Iteration 5 Input 0.154 Group # 3 -- f: 0.444, s: 0.607, d: 0.710, N: 0.602 Log likelihood = -518.983 Significance = 0.000 Run # 5, 2 cells: Convergence at Iteration 3 Input 0.161 Group # 4 -- C: 0.515, V: 0.480 Log likelihood = -534.242 Significance = 0.372 Run # 6, 3 cells: Convergence at Iteration 4 Input 0.160 Group # 5 -- z: 0.505, c: 0.555, x: 0.388 Log likelihood = -533.230 Significance = 0.243
82
Add Group # 2 with factors eqk ---------- Level # 2 ---------- Run # 7, 6 cells: Convergence at Iteration 5 Input 0.145 Group # 1 -- 2: 0.536, 1: 0.490 Group # 2 -- e: 0.445, q: 0.896, k: 0.590 Log likelihood = -483.250 Significance = 0.387 Run # 8, 9 cells: No Convergence at Iteration 20 Input 0.144 Group # 2 -- e: 0.481, q: 0.903, k: 0.235 Group # 3 -- f: 0.453, s: 0.557, d: 0.491, N: 0.830 Log likelihood = -479.949 Significance = 0.063 Run # 9, 6 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.446, q: 0.902, k: 0.563 Group # 4 -- C: 0.527, V: 0.464 Log likelihood = -482.605 Significance = 0.158 Run # 10, 5 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.446, q: 0.898, k: 0.574 Group # 5 -- z: 0.505, c: 0.510, x: 0.430 Log likelihood = -483.267 Significance = 0.684 No remaining groups significant Groups selected while stepping up: 2 Best stepping up run: #3 --------------------------------------------- # Stepping down: # Stepping down: ---------- Level # 5 ---------- Run # 11, 40 cells: No Convergence at Iteration 20 Input 0.143 Group # 1 -- 2: 0.530, 1: 0.492 Group # 2 -- e: 0.454, q: 0.892, k: 0.511 Group # 3 -- f: 0.447, s: 0.544, d: 0.486, N: 0.863 Group # 4 -- C: 0.524, V: 0.468 Group # 5 -- z: 0.543, c: 0.172, x: 0.446 Log likelihood = -477.718 ---------- Level # 4 ---------- Run # 12, 23 cells: No Convergence at Iteration 20 Input 0.144 Group # 2 -- e: 0.454, q: 0.894, k: 0.505 Group # 3 -- f: 0.447, s: 0.547, d: 0.486, N: 0.861 Group # 4 -- C: 0.525, V: 0.467
83
Group # 5 -- z: 0.543, c: 0.173, x: 0.443 Log likelihood = -478.023 Significance = 0.450 Run # 13, 26 cells: No Convergence at Iteration 20 Input 0.152 Group # 1 -- 2: 0.566, 1: 0.483 Group # 3 -- f: 0.409, s: 0.563, d: 0.675, N: 0.887 Group # 4 -- C: 0.512, V: 0.484 Group # 5 -- z: 0.550, c: 0.148, x: 0.418 Log likelihood = -512.328 Significance = 0.000 Run # 14, 18 cells: Convergence at Iteration 6 Input 0.144 Group # 1 -- 2: 0.530, 1: 0.492 Group # 2 -- e: 0.446, q: 0.898, k: 0.577 Group # 4 -- C: 0.526, V: 0.465 Group # 5 -- z: 0.507, c: 0.494, x: 0.432 Log likelihood = -481.920 Significance = 0.041 Run # 15, 23 cells: No Convergence at Iteration 20 Input 0.144 Group # 1 -- 2: 0.534, 1: 0.491 Group # 2 -- e: 0.454, q: 0.889, k: 0.513 Group # 3 -- f: 0.446, s: 0.550, d: 0.486, N: 0.865 Group # 5 -- z: 0.542, c: 0.179, x: 0.448 Log likelihood = -478.572 Significance = 0.194 Run # 16, 28 cells: No Convergence at Iteration 20 Input 0.144 Group # 1 -- 2: 0.534, 1: 0.491 Group # 2 -- e: 0.481, q: 0.903, k: 0.228 Group # 3 -- f: 0.453, s: 0.549, d: 0.490, N: 0.830 Group # 4 -- C: 0.523, V: 0.469 Log likelihood = -478.705 Significance = 0.385 Cut Group # 1 with factors 21 ---------- Level # 3 ---------- Run # 17, 15 cells: No Convergence at Iteration 20 Input 0.153 Group # 3 -- f: 0.409, s: 0.569, d: 0.680, N: 0.881 Group # 4 -- C: 0.514, V: 0.481 Group # 5 -- z: 0.551, c: 0.147, x: 0.412 Log likelihood = -513.956 Significance = 0.000 Run # 18, 10 cells: Convergence at Iteration 5 Input 0.144 Group # 2 -- e: 0.446, q: 0.901, k: 0.571 Group # 4 -- C: 0.527, V: 0.464 Group # 5 -- z: 0.507, c: 0.492, x: 0.429 Log likelihood = -482.192 Significance = 0.042 Run # 19, 13 cells: No Convergence at Iteration 20
84
Input 0.144 Group # 2 -- e: 0.454, q: 0.892, k: 0.507 Group # 3 -- f: 0.447, s: 0.553, d: 0.486, N: 0.863 Group # 5 -- z: 0.542, c: 0.181, x: 0.445 Log likelihood = -478.963 Significance = 0.178 Run # 20, 16 cells: No Convergence at Iteration 20 Input 0.144 Group # 2 -- e: 0.481, q: 0.906, k: 0.226 Group # 3 -- f: 0.454, s: 0.551, d: 0.491, N: 0.827 Group # 4 -- C: 0.524, V: 0.468 Log likelihood = -479.055 Significance = 0.367 Cut Group # 5 with factors zcx ---------- Level # 2 ---------- Run # 21, 8 cells: Convergence at Iteration 5 Input 0.153 Group # 3 -- f: 0.445, s: 0.604, d: 0.712, N: 0.594 Group # 4 -- C: 0.512, V: 0.484 Log likelihood = -518.745 Significance = 0.000 Run # 22, 6 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.446, q: 0.902, k: 0.563 Group # 4 -- C: 0.527, V: 0.464 Log likelihood = -482.605 Significance = 0.073 Run # 23, 9 cells: No Convergence at Iteration 20 Input 0.144 Group # 2 -- e: 0.481, q: 0.903, k: 0.235 Group # 3 -- f: 0.453, s: 0.557, d: 0.491, N: 0.830 Log likelihood = -479.949 Significance = 0.186 Cut Group # 4 with factors CV ---------- Level # 1 ---------- Run # 24, 4 cells: Convergence at Iteration 5 Input 0.154 Group # 3 -- f: 0.444, s: 0.607, d: 0.710, N: 0.602 Log likelihood = -518.983 Significance = 0.000 Run # 25, 3 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.445, q: 0.899, k: 0.583 Log likelihood = -483.652 Significance = 0.063 Cut Group # 3 with factors fsdN ---------- Level # 0 ---------- Run # 26, 1 cells: Convergence at Iteration 2
85
Input 0.161 Log likelihood = -534.667 Significance = 0.000 All remaining groups significant Groups eliminated while stepping down: 1 5 4 3 Best stepping up run: #3 Best stepping down run: #25
86
87
APPENDIX D
ONE-STEP BINOMIAL: SECOND CODING
(PHONOLOGICAL ENVIRONMENT)
Binomial Varbrul, 1 step ======================== Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. - One-level analysis only:One-level binomial analysis: Run # 1, 8 cells: Convergence at Iteration 5 Input 0.153 Group Factor Weight App/Total Input&Weight 1: f 0.445 0.13 0.13 s 0.604 0.22 0.22 d 0.712 0.31 0.31 N 0.594 0.22 0.21 2: C 0.512 0.17 0.16 V 0.484 0.15 0.15 Cell Total App'ns Expected Error sV 20 5 4.112 0.242 sC 44 9 9.886 0.102 fV 417 42 50.044 1.469 fC 483 72 63.991 1.155 dV 74 27 21.907 1.682 dC 72 18 23.051 1.628 NV 10 4 1.989 2.536 NC 92 18 20.019 0.260 Total Chi-square = 9.0759 Chi-square/cell = 1.1345 Log likelihood = -518.745
88
89
APPENDIX E
STEP-UP/STEP-DOWN ANALYSIS: SECOND CODING
(PHONOLOGICAL ENVIRONMENT)
Binomial Varbrul ================ Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. Threshold, step-up/down: 0.050001 # Stepping up: # Stepping up: ---------- Level # 0 ---------- Run # 1, 1 cells: Convergence at Iteration 2 Input 0.161 Log likelihood = -534.667 ---------- Level # 1 ---------- Run # 2, 4 cells: Convergence at Iteration 5 Input 0.154 Group # 1 -- f: 0.444, s: 0.607, d: 0.710, N: 0.602 Log likelihood = -518.983 Significance = 0.000 Run # 3, 2 cells: Convergence at Iteration 3 Input 0.161 Group # 2 -- C: 0.515, V: 0.480 Log likelihood = -534.242 Significance = 0.372 Add Group # 1 with factors fsdN ---------- Level # 2 ---------- Run # 4, 8 cells: Convergence at Iteration 5 Input 0.153 Group # 1 -- f: 0.445, s: 0.604, d: 0.712, N: 0.594 Group # 2 -- C: 0.512, V: 0.484 Log likelihood = -518.745 Significance = 0.493 No remaining groups significant Groups selected while stepping up: 1 Best stepping up run: #2 --------------------------------------------- # Stepping down:
90
# Stepping down: ---------- Level # 2 ---------- Run # 5, 8 cells: Convergence at Iteration 5 Input 0.153 Group # 1 -- f: 0.445, s: 0.604, d: 0.712, N: 0.594 Group # 2 -- C: 0.512, V: 0.484 Log likelihood = -518.745 ---------- Level # 1 ---------- Run # 6, 2 cells: Convergence at Iteration 3 Input 0.161 Group # 2 -- C: 0.515, V: 0.480 Log likelihood = -534.242 Significance = 0.000 Run # 7, 4 cells: Convergence at Iteration 5 Input 0.154 Group # 1 -- f: 0.444, s: 0.607, d: 0.710, N: 0.602 Log likelihood = -518.983 Significance = 0.493 Cut Group # 2 with factors CV
91
92
APPENDIX F
ONE-STEP BINOMIAL ANALYSIS: SECOND CODING
(MORPHOSYNTACTIC ENVIRONMENT)
Binomial Varbrul, 1 step ======================== Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. - One-level analysis only:One-level binomial analysis: Run # 1, 10 cells: Convergence at Iteration 5 Input 0.145 Group Factor Weight App/Total Input&Weight 1: 2 0.535 0.21 0.16 1 0.491 0.15 0.14 2: e 0.446 0.12 0.12 q 0.895 0.60 0.59 k 0.580 0.19 0.19 3: z 0.505 0.16 0.15 c 0.511 0.19 0.15 x 0.433 0.11 0.11 Cell Total App'ns Expected Error 2qz 33 22 20.776 0.195 2qx 2 1 1.120 0.029 2kc 5 3 1.096 4.239 2ez 203 25 28.012 0.376 2ex 10 1 1.070 0.005 1qz 52 29 30.561 0.193 1qx 1 1 0.516 0.938 1kc 94 16 17.908 0.251 1ez 732 90 86.639 0.148 1ex 80 7 7.303 0.014 Total Chi-square = 6.3867 Chi-square/cell = 0.6387 Log likelihood = -482.901
93
94
APPENDIX G
STEP-UP/STEP-DOWN ANALYSIS: SECOND CODING
(MORPHOSYNTACTIC ENVIRONMENT)
Binomial Varbrul ================ Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. Threshold, step-up/down: 0.050001 # Stepping up: # Stepping up: ---------- Level # 0 ---------- Run # 1, 1 cells: Convergence at Iteration 2 Input 0.161 Log likelihood = -534.667 ---------- Level # 1 ---------- Run # 2, 2 cells: Convergence at Iteration 4 Input 0.160 Group # 1 -- 2: 0.576, 1: 0.480 Log likelihood = -532.419 Significance = 0.037 Run # 3, 3 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.445, q: 0.899, k: 0.583 Log likelihood = -483.652 Significance = 0.000 Run # 4, 3 cells: Convergence at Iteration 4 Input 0.160 Group # 3 -- z: 0.505, c: 0.555, x: 0.388 Log likelihood = -533.230 Significance = 0.243 Add Group # 2 with factors eqk ---------- Level # 2 ---------- Run # 5, 6 cells: Convergence at Iteration 5 Input 0.145 Group # 1 -- 2: 0.536, 1: 0.490 Group # 2 -- e: 0.445, q: 0.896, k: 0.590 Log likelihood = -483.250 Significance = 0.387 Run # 6, 5 cells:
95
Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.446, q: 0.898, k: 0.574 Group # 3 -- z: 0.505, c: 0.510, x: 0.430 Log likelihood = -483.267 Significance = 0.684 No remaining groups significant Groups selected while stepping up: 2 Best stepping up run: #3 --------------------------------------------- # Stepping down: # Stepping down: ---------- Level # 3 ---------- Run # 7, 10 cells: Convergence at Iteration 5 Input 0.145 Group # 1 -- 2: 0.535, 1: 0.491 Group # 2 -- e: 0.446, q: 0.895, k: 0.580 Group # 3 -- z: 0.505, c: 0.511, x: 0.433 Log likelihood = -482.901 ---------- Level # 2 ---------- Run # 8, 5 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.446, q: 0.898, k: 0.574 Group # 3 -- z: 0.505, c: 0.510, x: 0.430 Log likelihood = -483.267 Significance = 0.410 Run # 9, 6 cells: Convergence at Iteration 4 Input 0.159 Group # 1 -- 2: 0.578, 1: 0.479 Group # 3 -- z: 0.503, c: 0.572, x: 0.396 Log likelihood = -530.912 Significance = 0.000 Run # 10, 6 cells: Convergence at Iteration 5 Input 0.145 Group # 1 -- 2: 0.536, 1: 0.490 Group # 2 -- e: 0.445, q: 0.896, k: 0.590 Log likelihood = -483.250 Significance = 0.706 Cut Group # 3 with factors zcx ---------- Level # 1 ---------- Run # 11, 3 cells: Convergence at Iteration 5 Input 0.145 Group # 2 -- e: 0.445, q: 0.899, k: 0.583 Log likelihood = -483.652 Significance = 0.387 Run # 12, 2 cells: Convergence at Iteration 4 Input 0.160
96
Group # 1 -- 2: 0.576, 1: 0.480 Log likelihood = -532.419 Significance = 0.000 Cut Group # 1 with factors 21 ---------- Level # 0 ---------- Run # 13, 1 cells: Convergence at Iteration 2 Input 0.161 Log likelihood = -534.667 Significance = 0.000 All remaining groups significant Groups eliminated while stepping down: 3 1 Best stepping up run: #3 Best stepping down run: #11
97
98
APPENDIX H
GOLDVARB 2001 CODING KEY
(SELECTED DATA)
Factor group 1: Subject type q = Noun phrase w = Pronoun e = [- overt] subject
Factor group 2: Preceding phonological environment
a = Vowel (or vowel sound) s = Consonant d = Nasal vowel f = No preceding phonological environment
Factor group 3: Following phonological environment
g = Vowel (or vowel sound) h = Consonant
Factor group 4: Sentence type z = Declarative x = Interrogative c = Imperative
Factor group 5: Discourse style v = Ludic m =Explanatory b = Emphatic n = Proverbial
99
100
APPENDIX I
GOLDVARB 2001 RESULTS FOR SELECTED DATA:
PHOLOGICAL ENVIRONMENT
Binomial Varbrul, 1 step ======================== Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. - One-level analysis only:One-level binomial analysis: Run # 1, 8 cells: Convergence at Iteration 5 Input 0.142 Group Factor Weight App/Total Input&Weight 1: a 0.404 0.10 0.10 f 0.577 0.21 0.18 s 0.883 0.56 0.56 d 0.804 0.40 0.40 2: g 0.436 0.13 0.11 h 0.552 0.20 0.17 Cell Total App'ns Expected Error sh 10 5 6.059 0.470 sg 8 5 3.934 0.568 fh 21 4 4.563 0.089 fg 3 1 0.446 0.808 dh 5 2 2.275 0.061 dg 5 2 1.722 0.068 ah 75 11 9.090 0.457 ag 74 4 5.911 0.671 Total Chi-square = 3.1910 Chi-square/cell = 0.3989 Log likelihood = -79.451 Binomial Varbrul ================ Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. Threshold, step-up/down: 0.050001 # Stepping up: # Stepping up: ---------- Level # 0 ----------
101
Run # 1, 1 cells: Convergence at Iteration 2 Input 0.169 Log likelihood = -91.363 ---------- Level # 1 ---------- Run # 2, 4 cells: Convergence at Iteration 5 Input 0.144 Group # 1 -- a: 0.400, f: 0.610, s: 0.881, d: 0.798 Log likelihood = -80.034 Significance = 0.000 Run # 3, 2 cells: Convergence at Iteration 4 Input 0.167 Group # 2 -- g: 0.435, h: 0.553 Log likelihood = -90.607 Significance = 0.223 Add Group # 1 with factors afsd ---------- Level # 2 ---------- Run # 4, 8 cells: Convergence at Iteration 5 Input 0.142 Group # 1 -- a: 0.404, f: 0.577, s: 0.883, d: 0.804 Group # 2 -- g: 0.436, h: 0.552 Log likelihood = -79.451 Significance = 0.284 No remaining groups significant Groups selected while stepping up: 1 Best stepping up run: #2 --------------------------------------------- # Stepping down: # Stepping down: ---------- Level # 2 ---------- Run # 5, 8 cells: Convergence at Iteration 5 Input 0.142 Group # 1 -- a: 0.404, f: 0.577, s: 0.883, d: 0.804 Group # 2 -- g: 0.436, h: 0.552 Log likelihood = -79.451 ---------- Level # 1 ---------- Run # 6, 2 cells: Convergence at Iteration 4 Input 0.167 Group # 2 -- g: 0.435, h: 0.553 Log likelihood = -90.607 Significance = 0.000 Run # 7, 4 cells: Convergence at Iteration 5 Input 0.144 Group # 1 -- a: 0.400, f: 0.610, s: 0.881, d: 0.798 Log likelihood = -80.034 Significance = 0.284
102
Cut Group # 2 with factors gh ---------- Level # 0 ---------- Run # 8, 1 cells: Convergence at Iteration 2 Input 0.169 Log likelihood = -91.363 Significance = 0.000 All remaining groups significant Groups eliminated while stepping down: 2 Best stepping up run: #2 Best stepping down run: #7
103
104
APPENDIX J
GOLDVARB 2001 RESULTS FOR SELECTED DATA:
MORPHOSYNTACTIC ENVIRONMENT AND DISCOURSE STYLE
Binomial Varbrul, 1 step ======================== Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. - One-level analysis only:One-level binomial analysis: Run # 1, 15 cells: Convergence at Iteration 12 Input 0.051 Group Factor Weight App/Total Input&Weight 1: w 0.434 0.12 0.04 e 0.421 0.17 0.04 q 0.981 0.77 0.74 2: z 0.505 0.18 0.05 x 0.439 0.07 0.04 c 0.500 0.18 0.05 3: v 0.917 0.39 0.37 m 0.140 0.02 0.01 b 0.853 0.23 0.24 n 0.947 0.50 0.49 Cell Total App'ns Expected Error wzv 39 14 12.400 0.303 wzn 7 3 3.018 0.000 wzm 90 0 0.616 0.621 wzb 15 2 2.965 0.392 wxv 3 1 0.789 0.077 wxm 10 0 0.052 0.053 wxb 1 0 0.159 0.188 qzv 7 6 6.788 3.012 qzn 1 1 0.981 0.019 qzm 4 2 1.284 0.588 qzb 1 1 0.944 0.059 ezm 1 0 0.006 0.007 ecv 10 2 3.022 0.495 ecm 7 0 0.045 0.045 ecb 5 2 0.931 1.507 Total Chi-square = 7.3658 Chi-square/cell = 0.4911 Log likelihood = -55.390
105
Binomial Varbrul ================ Name of cell file: Untitled.cel Using fast, less accurate method. Averaging by weighting factors. Threshold, step-up/down: 0.050001 # Stepping up: # Stepping up: ---------- Level # 0 ---------- Run # 1, 1 cells: Convergence at Iteration 2 Input 0.169 Log likelihood = -91.363 ---------- Level # 1 ---------- Run # 2, 3 cells: Convergence at Iteration 5 Input 0.151 Group # 1 -- w: 0.437, e: 0.542, q: 0.949 Log likelihood = -78.589 Significance = 0.000 Run # 3, 3 cells: Convergence at Iteration 5 Input 0.166 Group # 2 -- z: 0.517, x: 0.279, c: 0.527 Log likelihood = -90.742 Significance = 0.543 Run # 4, 4 cells: Convergence at Iteration 6 Input 0.076 Group # 3 -- v: 0.886, m: 0.181, b: 0.781, n: 0.924 Log likelihood = -66.821 Significance = 0.000 Add Group # 3 with factors vmbn ---------- Level # 2 ---------- Run # 5, 11 cells: Convergence at Iteration 12 Input 0.052 Group # 1 -- w: 0.433, e: 0.419, q: 0.982 Group # 3 -- v: 0.917, m: 0.140, b: 0.854, n: 0.948 Log likelihood = -55.417 Significance = 0.000 Run # 6, 10 cells: Convergence at Iteration 6 Input 0.075 Group # 2 -- z: 0.526, x: 0.348, c: 0.407 Group # 3 -- v: 0.888, m: 0.179, b: 0.788, n: 0.918 Log likelihood = -66.345 Significance = 0.629 Add Group # 1 with factors weq ---------- Level # 3 ----------
106
Run # 7, 15 cells: Convergence at Iteration 12 Input 0.051 Group # 1 -- w: 0.434, e: 0.421, q: 0.981 Group # 2 -- z: 0.505, x: 0.439, c: 0.500 Group # 3 -- v: 0.917, m: 0.140, b: 0.853, n: 0.947 Log likelihood = -55.390 Significance = 0.974 No remaining groups significant Groups selected while stepping up: 3 1 Best stepping up run: #5 --------------------------------------------- # Stepping down: # Stepping down: ---------- Level # 3 ---------- Run # 8, 15 cells: Convergence at Iteration 12 Input 0.051 Group # 1 -- w: 0.434, e: 0.421, q: 0.981 Group # 2 -- z: 0.505, x: 0.439, c: 0.500 Group # 3 -- v: 0.917, m: 0.140, b: 0.853, n: 0.947 Log likelihood = -55.390 ---------- Level # 2 ---------- Run # 9, 10 cells: Convergence at Iteration 6 Input 0.075 Group # 2 -- z: 0.526, x: 0.348, c: 0.407 Group # 3 -- v: 0.888, m: 0.179, b: 0.788, n: 0.918 Log likelihood = -66.345 Significance = 0.000 Run # 10, 11 cells: Convergence at Iteration 12 Input 0.052 Group # 1 -- w: 0.433, e: 0.419, q: 0.982 Group # 3 -- v: 0.917, m: 0.140, b: 0.854, n: 0.948 Log likelihood = -55.417 Significance = 0.974 Run # 11, 5 cells: No Convergence at Iteration 20 Input 0.150 Group # 1 -- w: 0.456, e: 0.399, q: 0.951 Group # 2 -- z: 0.493, x: 0.343, c: 0.654 Log likelihood = -78.299 Significance = 0.000 Cut Group # 2 with factors zxc ---------- Level # 1 ---------- Run # 12, 4 cells: Convergence at Iteration 6 Input 0.076 Group # 3 -- v: 0.886, m: 0.181, b: 0.781, n: 0.924 Log likelihood = -66.821 Significance = 0.000 Run # 13, 3 cells:
107
Convergence at Iteration 5 Input 0.151 Group # 1 -- w: 0.437, e: 0.542, q: 0.949 Log likelihood = -78.589 Significance = 0.000 All remaining groups significant Groups eliminated while stepping down: 2 Best stepping up run: #5 Best stepping down run: #10
108
REFERENCES
Anis, J. (Ed.). (1999). Internet communication et langue française [Internet communication
and the French language]. Paris: Hermès Science.
Armstrong, N. (2002). Variable deletion of French ne: A cross-stylistic perspective.
Language Sciences, 24, 153-173.
Armstrong, N., & Smith, A. (2002). The influence of linguistic and social factors on the
recent decline of French ne. Journal of French Language Studies, 12, 23-41.
Ashby, W. (1981). The loss of the negative particle ne in French: A syntactic change in
progress. Language, 57, 674-687.
Ashby, W. (2001). Un nouveau regard sur la chute du ne en français parlé tourangeau:
S'agit-il d'un changement en cours? [A new look at the loss of ne in spoken Tours
French: A change in progess?] Journal of French Language Studies, 11, 1-22.
Bell, A. (1984). Language style as audience design. Language in Society, 13, 145-204.
Bell, A. (2001). Back in style: Reworking audience design. In P. Eckert & J. R. Rickford
(Eds.) Style and sociolinguistic variation (pp. 139-169). Cambridge, England:
Cambridge University Press.
Biber, D., & Finegan, E. (1994). Sociolinguistic perspectives on register. Oxford, England:
Oxford University Press.
Blanche-Benveniste, C. (1997). La notion de variation syntaxique dans la langue parlée [The
notion of syntactic variation in speech]. Langue Française, 115, 19-29.
Brunot, F. (1966). Histoire de la langue française: Des origines à nos jours [History of the
French language: From the origins to our days]. Paris: Armand Colin.
Butler, C. (1985). Statistics in linguistics. Oxford, England: Basil Blackwell Ltd.
109
Chambers, J. K. (2003). Sociolinguistic theory: Linguistic variation and its significance.
Oxford, England: Blackwell Publishing.
Cohen, M. (1967). Histoire d'une langue: Le français (des lointaines origines à nos jours)
[History of a language: French (from the origins to our days)] (3rd ed.). Paris:
Editions Sociales.
Collot, M., & Belmore, N. (1996). Electronic language: A new variety of English. In S.
Herring (Ed.), Computer-mediated communication: Linguistic, social and cross-
cultural perspectives (pp. 13-28). Philidelphia: John Benjamins.
Coveney, A. (1996). Variability in spoken French: A sociolinguistic study of interrogation
and negation. Exeter, England: Elm Bank.
Dauzat, A. (1954). Le génie de la langue française [The genius of the French language].
Paris: Payot.
Dauzat, A. (1967). Tableau de la langue française [A painting of the French language]. Paris:
Payot.
Dejond, A. (2002). La cyberlangue française [The French cyberlanguage]. Tournai, France:
La Renaissance du Livre.
Eckert, P., & Rickford, J. (Eds.). (2001). Style and sociolinguistic variation. Cambridge,
England: Cambridge University Press.
Ewert, A. (1969). The French language. London: Faber & Faber Limited. (Original work
published in 1933).
Fischer, J. (1958). Social influence of a linguistic variant. Word, 14, 47-56.
Garcia, A. C., & Jacobs, J. B. (1999). The eyes of the beholder: Understanding the turn-
taking system in quasi-synchronous computer-mediated communication. Research on
Language and Social Interaction, 32, 337-367.
110
Grevisse, M. (1993). Le bon usage: Grammaire française [The right usage: French
grammar]. 13th ed. Paris: Duculot.
Halliday, M. A. K. (1976). A brief sketch of systemic grammar. In G. R. Kress (Ed.),
Halliday: System and function in language (pp. 3-6). Oxford, England: Oxford
University Press.
Hansen, A., & Malderez, I. (2004). Le ne de négation en région parisienne: Une étude en
temps réel [The ne of negation in the Paris area: A real-time study]. Langage &
Société, 107, 5-30.
Hatch, Evelyn, & Lazaraton, Anne. (1991). The research manual: Design and statistics for
applied linguistics. New York: Newbury House.
Hearne, G., Mandeville, T., & Anthony, D. (1998). The communication superhighway:
Social and economic change in the Digital Age. St. Leonards: Allen & Unwin.
Herring, S. (1999). Interactional coherence in CMC. Journal of Computer-Mediated
Communication, 4. Retrieved November 25, 2006, from
http://jcmc.indiana.edu/vol4/issue4/herring.html/
Herring, S. C. (2001). Computer-mediated discourse. In D. Schiffrin, D. Tannen, & H.
Hamilton (Eds.), The handbook of discourse analysis (pp. 612-634). Oxford, England:
Blackwell
Hymes, D. (1974). Foundations of sociolinguistics. Philadelphia: University of Pennsylvania
Press.
Kayne, R. (1983). Connectedness. Linguistic Inquiry, 14, 223-249.
Labov, W. (1972). Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.
Lavandera, B. (1978). Where does the sociolinguistic variable stop? Language in Society, 7,
171-182.
111
Lewis, D. M. (2005). Arguing in English and French asynchronous online discussion.
Journal of Pragmatics, 37, 1,801-1,818.
Milroy, L. (1987). Observing and analysing natural language: A critical account of
sociolinguistic method. Oxford, England: Basil Blackwell.
Mosco, V. (2004). The digital sublime: Myth, power, and cyberspace. Cambridge: MIT.
Peeters, B. (2006). "Nous on vous tu(e)": La guerre (pacifique) des pronoms personnels
["Nous on vous tu(e)": The (peaceful) war of personal pronouns]. Zeitschrift für
romanische Philologie, 122, 201-220.
Peretz-Juillard, C. (1977). Les voyelles orales à Paris dans les dynamiques des âges de la
société [Oral vowels in Paris in the dynamics of ages of society]. Unpublished
doctoral dissertation, Université de Paris V.
Pierozak, I. (2003a). Le "français tchaté": Un objet à géométrie variable? ["Chatted French":
An object of variable geometry?] Langage et Société,104, 123-144.
Pierozak, I. (2003b). La variation à la marge sur Internet: Pseudos et (re)présentations
[Variation on the edge on the Internet: Pseudos and (re)presentations]. Cahiers du
Français Comptemporain, 8, 195-222.
Pierozak, I. (2003c). Le français tchaté. Une étude en trois dimensions—sociolinguistique,
syntaxique et graphique—d'usages IRC [Chatted French. A study in three
dimensions—sociolinguistic, syntactic, and graphic—of IRC usages]. Unpublished
doctoral dissertation, Université d'Aix-Marseille I.
Pohl, J. (1975). L'ommission de ne dans le français parlé contemporain [The omission of ne
in contemporary spoken French]. Le Français dans le Monde, 111, 17-23.
Pope, M. K. (1961). From Latin to Modern French with especial consideration of Anglo-
norman: Phonology and morphology. Manchester, England: Manchester University
Press. (Original work published in 1934).
112
Rickard, P. (1989). A history of the French language. 2nd ed. London: Unwin Hyman.
Rohlfs, G. (1970). From Vulgar Latin to Old French: An introduction to the study of the Old
French language (V. Almazan & L. McCarthy, Trans.). Detroit: Wayne State
University Press. (Original work published in 1960).
Sancier-Chateau, A. (1993). Introduction à la langue du XVIIe siècle: Syntaxe [Introduction
to the language of the XVIIe century: Syntax]. (C. Thomasset, Dir.). Paris: Nathan.
Sankoff, G. (1972). Above and beyond phonology in variable rules. In C.-J. Baily & R. W.
Shuy (Eds.), New ways of analyzing variation in English. Washington, D. C.:
Georgetown University Press.
Sankoff, G., & Vincent, D. (1977). L'emploi productif de ne dans le français parlé à
Montréal [The productive use of ne in Montreal spoken French]. Le Français
Moderne, 45, 243-256.
Sankoff, G. (1980). The social life of language. Philadelphia: University of Pennsylvania
Press.
Sax, K. (2003). Acquisition of stylistic variation by American learners of French.
Dissertation Abstracts International, 65 (02), 491A. (UMI No. 200505889)
Tagliamonte, S. A. (2006). Analysing sociolinguistic variation. Cambridge, England:
Cambridge University Press.
Thorne, S. (1999). An activity theoretical analysis of foreign language electronic
discourse. Dissertation Abstracts International, 61 (03), 920A. (UMI No. 200107598)
Trudgill, P. (Ed.). (1984). Applied sociolinguistics. London: Academic Press Inc.
van Compernolle, R. (2006, May). The variable use of ne in public French-language chat.
Paper presented at CMT 2006, Bordeaux, France.
113
van Compernolle, R., & Williams, L. (in press). De l'oral à l'électronique: la variation
orthographique comme ressource sociostylistique et pragmatique dans le discours
électronique. Glottopol.
van Lier, L. (1988). The classroom and the language learner: Ethnography and second-
language classroom research. London: Longman Publishing Group.
Werry, C. (1996). Linguistic and interactional features of Internet Relay Chat. In S. Herring
(Ed.), Computer-mediated communication: Linguistic, social and cross-cultural
perspectives (pp. 47-64). Philadelphia: John Benjamins.
Whittle, D. B. (1997). Cyberspace: The human dimension. New York: Freeman.
Williams, L. (2003). The nature and complexities of chat discourse: A qualitative case study
of multi-level learners of French in an electronic environment. Dissertation Abstracts
International, 64 (09), 3,275A. (UMI No. 200500146)
Williams, L. (2006, May). Orthographic and sociopragmatic norms and variation in French-
language web forums, moderated chat, and non-moderated chat. Paper presented at
CMT 2006, Bordeaux, France.
Williams, L., & van Compernolle, R. (2007). Second-person pronoun use in on-line French-
language chat environments. The French Review, 80, 804-820.
Wolfram, W. (1991). The linguistic variable: Fact and fantasy, American Speech, 66, 22-32.
114